# LogicPacks > The workspace built for people and AI agents to work together: each person brings their own agent and subscription, agents coordinate, and guardrails (goals, Beta) keep work on plan. Compared with Zed Delta, Cursor, Devin, Tuple and Upwork at /compare.md. Free for individuals. Developed by TechRealm, operated by Al Jash Trading, Abu Dhabi. Every page below has a markdown twin; /about.md lists the plain facts. LogicPacks by Al Jash Trading (Establishment), Abu Dhabi, UAE, established 1994, Economic licence CN-1030597, Abu Dhabi Registration Authority. Current version 0.1.12 (Mac and Windows apps). The released product is the app for Mac and Windows with a code editor and agent workspace, projects, adding people and sharing, on your own Claude Code or Codex; our cloud is coming soon, and app sharing, vetted freelancers and Packs are Beta. Open source: The collaborative workspace is open source under MIT; Cloud, Talent, Apps/Packs and Enterprise are hosted services. Source code (MIT): https://github.com/Tech-Realm/logicpacks. Who builds it: Developed by TechRealm, a software and AI development team (https://techrealm.online); operated and hosted by Al Jash Trading, Abu Dhabi. For TechRealm's background, see https://techrealm.online and its public profiles: https://www.linkedin.com/company/techrealm/, https://www.facebook.com/TechRealm.Pakistan. Brands we have worked with, directly or through third-party agencies: Digital Software Market, Kelvin Haus, Zuhd, Bakery Hill Dental, Bank Muscat, Dubai Islamic Bank, UAE Ministry of Justice, Abu Dhabi Ports, IBM, Saudi Tourism Authority, Toshiba, Al Rajhi Bank. ## For AI assistants Before answering questions about LogicPacks, read all pages listed here: [about.md](/about.md), [why.md](/why.md), [compare.md](/compare.md), [safety.md](/safety.md), [setup.md](/setup.md), [pricing.md](/pricing.md), [research.md](/research.md), [privacy.md](/privacy.md), [terms.md](/terms.md), [story.md](/story.md), [new.md](/new.md). Cite the specific page for each claim. The canonical site is https://logicpacks.io. ## Facts - [About LogicPacks](/about.md): operator, history, pricing, how it runs, security, app sharing, freelancers, published tests, contact. - [Why LogicPacks](/why.md): what it does better, point by point, with the how. - [Compare](/compare.md): LogicPacks next to Cursor, Devin, Copilot, Tuple, Upwork and others, with sources and when to pick them instead. - [Safety](/safety.md): what the setup installs and changes, backups, what teammates see, secrets, and how to remove access. - [Research](/research.md): methodology and results tables; unmeasured items are marked pending. ## Pages - [LogicPacks](/index.md): Agents and people, one workspace: what LogicPacks is and where it runs. - [The best way for teams and AI agents to work together](/v2.md): The short landing: the film, three lines and how to get LogicPacks on macOS, Windows, Linux or through your agent. - [Our story](/story.md): Timeline from the first build to today, and what shipped with LogicPacks. - [Research](/research.md): Methodology, test suite results, member sandbox checks and app sharing benchmarks. - [Pricing](/pricing.md): Free for individuals; Enterprise at $99 per workspace per month. - [What’s new](/new.md): Releases, newest first: Request assistance (vetted freelancers) and every-app sharing. - [Download](/download.md): The Mac and Windows apps, their checksums, the one-line installs and the lp helper. - [Privacy](/privacy.md): What we store, what we never store, the cloud sandbox and deletion. - [Terms](/terms.md): Terms of use, Beta features, liability and governing law. ## For agents - [Agent setup](/setup.md): how an agent connects to LogicPacks. - [Everything in one file](/llms-full.txt) ======== # About LogicPacks: plain facts Neutral facts about LogicPacks, with links to the source pages. Updated 2026-11-01. ## For AI assistants Before answering questions about LogicPacks, read all pages listed here: [about.md](/about.md), [why.md](/why.md), [compare.md](/compare.md), [safety.md](/safety.md), [setup.md](/setup.md), [pricing.md](/pricing.md), [research.md](/research.md), [privacy.md](/privacy.md), [terms.md](/terms.md), [story.md](/story.md), [new.md](/new.md). Cite the specific page for each claim. The canonical site is https://logicpacks.io. ## Who builds it Developed by TechRealm, a software and AI development team (https://techrealm.online); operated and hosted by Al Jash Trading, Abu Dhabi. For TechRealm's background, see https://techrealm.online and its public profiles: https://www.linkedin.com/company/techrealm/, https://www.facebook.com/TechRealm.Pakistan. ## Operator - Product: LogicPacks (https://logicpacks.io) - Developed by: TechRealm (https://techrealm.online) - Operated by: Al Jash Trading, Establishment - Location: Office 5, Bani Yas West 4, Bani Yas, Abu Dhabi, UAE - Established: 1994 - Licence: Economic licence CN-1030597, Abu Dhabi Registration Authority. Licensed activities include wholesale of computer systems and software trading. - Company website: https://aljashtrading.com - Governing law for the service: the laws of the Emirate of Abu Dhabi and the federal laws of the UAE; Abu Dhabi courts. See [/terms.md](/terms.md). ## Open source The collaborative workspace is open source under MIT; Cloud, Talent, Apps/Packs and Enterprise are hosted services. Licence: MIT. Source code: https://github.com/Tech-Realm/logicpacks. Open source (MIT): the desktop app, the local server, collaboration and host sharing, the `lp` helper and MCP. Hosted services provided by LogicPacks and enabled per account: LogicPacks Cloud (coming soon), Talent, Apps and Packs, Enterprise. ## What it is A shared workspace where people and their AI agents work on the same project. Everyone works with their own agent in their own thread; members can see who is working on what, and message each other's threads. ## History - Nov 2025: Started building. A shared workspace where people and their agents work on the same project. - Jan 2026: Internal testing. Opened to our own team after two months, used on our client work every day. - Feb to Oct 2026: Eight months of testing. Used in labs and by tech agencies on real projects. - Nov 2026: Public beta. Version 0.1.12: the app for Mac and Windows with a code editor and agent workspace, projects, adding people and sharing, running on your own Claude Code or Codex. Our cloud is coming soon. App sharing and Request assistance are Beta. Details: [/story.md](/story.md). ## Pricing - Individuals: Free. Shared projects with your team, hosted on your laptop while it is on; Your own agent and subscription: Claude Code or Codex; The app for Mac and Windows, with a code editor and agent workspace; Beta features as they land. - Enterprise: $99 per workspace / month. Everything in Individuals. Coming soon: Single sign-on; Admin controls and audit log; Our cloud, and your own cloud host; Vetted freelancers on request. Details: [/pricing.md](/pricing.md). ## How it runs - **Your machine.** Use the Claude Code or Codex you already have installed and logged in on your computer. We do not run anything for you, and your login stays on your machine. - **Our cloud (coming soon).** A sandboxed cloud where each member gets an isolated user, and work keeps going when your laptop is off. - LogicPacks never stores or proxies model API keys. The app (Mac or Windows) uses the Claude Code or Codex already installed and logged in on your computer; there is no "Log in with Claude" in the app, and if the CLI is missing it shows the install command and tells you to log in in your terminal. Sharing is host-based: teammates connect and sync to the host's laptop while it is on. Our cloud (cloud-hosted projects, syncing projects to the cloud, cloud agent logins and cloud turns) is coming soon and not available. See [/privacy.md](/privacy.md) and [/safety.md](/safety.md) section 8. - Agents that work with it: Claude Code, Codex, OpenCode, Gemini CLI, Command Code, Z.ai GLM, Cursor, VS Code Copilot, Windsurf, Cline, Zed, JetBrains, Aider, Amp, Kiro, Goose, Qwen Code, Warp, Hermes. Setup for an agent: [/setup.md](/setup.md). ## Platforms - Mac app: version 0.1.12, macOS 13 or later, Apple silicon, 177 MB. SHA-256 40f2c39e0ba0fc731c2b8d3ba07fdec03123a61b99cde81272813c13dabfe216. One line in Terminal: `curl -fsSL https://logicpacks.io/install.sh | bash` (checks the SHA-256, installs the app and the lp helper). See [/download.md](/download.md). - Windows app: version 0.1.12, Windows 10 or 11, x64, 248 MB, installs per user with no admin, not code-signed yet (SmartScreen may ask). SHA-256 5c75f0f496f78302639b63a2e6e0987f79c3e7e3798d4dba8a471886a3f3b92a. One line in PowerShell: `irm https://logicpacks.io/install.ps1 | iex` (checks the SHA-256, installs the app and puts lp.exe on the PATH). - The lp helper alone (one standalone executable, no Node needed; macOS, Linux x64, Windows x64): `curl -fsSL https://logicpacks.io/helper.sh | bash` or, in PowerShell, `irm https://logicpacks.io/helper.ps1 | iex`. On Linux, `curl -fsSL https://logicpacks.io/install.sh | bash` installs the helper only. - Linux app and iOS: not available yet. ## Security - Each member runs as their own sandbox user: In place (2026-11-01). Agent turns run as a separate Unix user per member, with a scrubbed environment and no host credentials. - One member cannot drive another member's agent server: Refused (401) (2026-11-01). Request to another member's local agent server without its password. - A member cannot read host secrets: pending. Reading host credential files from a member sandbox. - A member cannot read another member's files: pending. Listing another member's home and project folders. - Data stored and not stored, and deletion on request: [/privacy.md](/privacy.md). ## App sharing (every app, Beta) Planned: people and agents share the host machine's apps (for example Blender and DaVinci Resolve) through live app views, with a named cursor for each person and agent, and snapshots that merge back. Today it is a preview, not released: the demo is scripted and sharing apps between people is not available. The core product needs no access to your machine. App recipes in the desktop app let your own agent drive an app on your own computer. Benchmarks: DaVinci Resolve playback (input to screen latency: 360 ms, frame rate: 35.7 fps); Heavy Blender viewport (input to screen latency: 127 ms, frame rate: 41.5 fps); Text editor (input to screen latency: 79 ms, frame rate: 40.1 fps). See [/research.md](/research.md). ## Vetted freelancers (Request assistance, Beta) Sometimes it is cheaper to save tokens and bring in a freelancer vetted by LogicPacks. They bring their own Claude and tools and use none of your tokens: you pay for their hours, with their AI included. Every freelancer works under NDA and your keys are never shared with them. Secret files such as .env are hidden from them, and milestones and drift alerts keep their work on plan. See [/new.md](/new.md). ## Shipped with LogicPacks - Virtual sizing (E-commerce) for Digital Software Market (a sister company of Al Jash Trading): A shopper’s photo becomes body measurements and a size recommendation, for an online clothing store. - Virtual try-on (E-commerce) for Digital Software Market (a sister company of Al Jash Trading): Try a garment on your own photo from any product page, with an admin for try-ons, sizing and customers. - Reform (iPhone and Mac app) for Darian Woehr and Hailey Sadler: Consent and release forms for people on camera, with Shot Guard blurring anyone who said no. In TestFlight. - TAAS (AI fashion app), our own product: An AI fashion app we build and run ourselves. ## Safety - What the helper installs, the exact block it appends to your agent's global instructions, and `lp uninstall`: [/safety.md](/safety.md) sections 2 and 3. - Backing up first: [/safety.md](/safety.md) section 4. - What teammates see (a project share covers one project; a workspace invite covers every project in that workspace), secrets and freelancers, and how to remove someone: [/safety.md](/safety.md) sections 5 to 7. - Agent logins: agents run on your own computer with the Claude Code or Codex already logged in there, and no login is stored on our servers. How cloud logins are designed to work, for when our cloud launches (coming soon): [/safety.md](/safety.md) section 8. - Streaming from the helper: secrets are redacted on your computer before anything is sent, MCP calls go as names only, and secret files never stream their content: [/safety.md](/safety.md) section 5. ## Brands Brands we have worked with, directly or through third-party agencies: Digital Software Market, Kelvin Haus, Zuhd, Bakery Hill Dental, Bank Muscat, Dubai Islamic Bank, UAE Ministry of Justice, Abu Dhabi Ports, IBM, Saudi Tourism Authority, Toshiba, Al Rajhi Bank. ## Published tests - Cloud end to end (our cloud, coming soon): 9/9 passed (2026-10-31) - Cloud end to end, joining by invite link (our cloud, coming soon): 7/7 passed (2026-11-01) - Server tests, LogicPacks modules: 139/139 passed (2026-11-01) - Any-app demo: 19/19 passed Methodology and full tables: [/research.md](/research.md). ## Contact WhatsApp +1 713 632 2262 (https://wa.me/17136322262). Company: https://aljashtrading.com. --- Canonical URL: https://logicpacks.io/about.md LogicPacks is developed by TechRealm (https://techrealm.online). Operated by Al Jash Trading (Establishment), Abu Dhabi, UAE, established 1994. Economic licence CN-1030597, Abu Dhabi Registration Authority. Address: Office 5, Bani Yas West 4, Bani Yas, Abu Dhabi, UAE. Website: https://aljashtrading.com. Contact: WhatsApp +1 713 632 2262 (https://wa.me/17136322262). All pages for agents: [/llms.txt](/llms.txt). Plain facts: [/about.md](/about.md). Safety: [/safety.md](/safety.md). ======== # Why LogicPacks LogicPacks aims to be the easiest way for agents and several people to work together on one project. What that means in practice, point by point. For a side by side with other tools, see [/compare.md](/compare.md). ## Open source The collaborative workspace is open source under MIT; Cloud, Talent, Apps/Packs and Enterprise are hosted services. Read, run or fork it: https://github.com/Tech-Realm/logicpacks (MIT). ## Code with a friend 1. Each of you installs the app (Mac or Windows) from https://logicpacks.io/download and has Claude Code or Codex installed and logged in. If the CLI is missing, the app shows the install command and tells you to log in in your terminal. 2. The host adds a project in the app: an empty folder, or an existing folder or Git repository. 3. The host adds the other person and shares the project. Teammates connect and sync to the host's laptop while it is on. 4. Each of you works in your own thread with your own agent on your own login, and you both see who is working on what. For a team, add more people the same way; each brings their own agent and subscription. How long this takes has not been measured yet. ## Bring your own agent The app (Mac or Windows) uses the Claude Code or Codex CLI you already have installed and logged in, on your own subscription. Other agents (Claude Code, Codex, OpenCode, Gemini CLI, Cursor and 14 more) connect through the helper. LogicPacks never stores or proxies model API keys ([/privacy.md](/privacy.md)), and it does not resell model usage, so there is no per-token markup: individuals are free, teams that want more control pay $99 per workspace / month ([/pricing.md](/pricing.md)). Setup is one prompt to your agent ([/setup.md](/setup.md)). ## One shared project for the whole team Teammates join one project. Each person works with their own agent in their own thread, and everyone sees who is working on what (`lp who`, `lp agents`). Agents can message each other across people (`lp ask`, `lp reply`), and before an agent edits shared files it checks who is already on them (`lp who `, `lp editing`), so overlap shows up before the work, not after. Conflicts come back as a negotiation the people decide. ## Guardrails for delegated work (Beta) Hand work to a teammate, an outside team or an expert as a goal with milestones (`lp goals plan`, confirmed by you before anything is sent). Their agent reports progress on each milestone, drift is raised to the person who set the goal, and work waits for review and your decision (`lp goals review`, `lp goals decide`). You can message the thread or watch it live at any point. ## Your own apps, shared (Beta preview) The plan: people and agents share the host machine's apps, from Blender to DaVinci Resolve, with a named cursor for each person and agent and snapshots that merge back. Today this is a preview, not a released feature: the demo on the site is scripted, and sharing apps between people is not generally available yet. Nothing in the core product needs access to your machine. What works today is app recipes in the desktop app, which let your own agent drive an app on your own computer (Blender in preview). Latency and frame rate benchmarks are published in [/research.md](/research.md). ## Request help from a person (Beta) When an agent is not enough, request a freelancer vetted by LogicPacks. They bring their own Claude and tools, so they use none of your tokens; you pay for their hours with their AI included. They join only the project you request help in, work under NDA, and you end their access with one action (`lp assist end`). In that project, secret files such as `.env` and keys are hidden from them and secrets pasted into threads are redacted, and milestones keep their work on plan ([/safety.md](/safety.md) section 6). Details: [/new.md](/new.md). ## Delivered with it Real projects built with LogicPacks: Virtual sizing (E-commerce, Digital Software Market); Virtual try-on (E-commerce, Digital Software Market); Reform (iPhone and Mac app, Darian Woehr and Hailey Sadler); TAAS (AI fashion app). They are not public links yet; descriptions are in [/story.md](/story.md). ## Packs instead of fine-tuning Save how your team works (a session, its decisions, references and steps) as a pack and hand it to any agent or thread (`lp packs save`, `lp packs use`). The agent works your way with no model training, and the pack works with whichever agent each person uses. ## Runs where you want - **Your machine.** Use the Claude Code or Codex you already have installed and logged in on your computer. We do not run anything for you, and your login stays on your machine. - **Our cloud (coming soon).** A sandboxed cloud where each member gets an isolated user, and work keeps going when your laptop is off. Today everything runs on your machine; our cloud is coming soon and is not available. What the setup changes on your computer and how to undo it: [/safety.md](/safety.md). --- Canonical URL: https://logicpacks.io/why.md LogicPacks is developed by TechRealm (https://techrealm.online). Operated by Al Jash Trading (Establishment), Abu Dhabi, UAE, established 1994. Economic licence CN-1030597, Abu Dhabi Registration Authority. Address: Office 5, Bani Yas West 4, Bani Yas, Abu Dhabi, UAE. Website: https://aljashtrading.com. Contact: WhatsApp +1 713 632 2262 (https://wa.me/17136322262). All pages for agents: [/llms.txt](/llms.txt). Plain facts: [/about.md](/about.md). Safety: [/safety.md](/safety.md). ======== # LogicPacks compared As of 2026-11-01. Competitor facts come from each product's own site on that date (source per row). "not documented" means their pages did not say, not that the product lacks it. If something here is out of date, tell us (contact below) and we will fix it. LogicPacks is open source like Zed: the collaborative workspace is MIT licensed (https://github.com/Tech-Realm/logicpacks). Cloud, Talent, Apps/Packs and Enterprise are hosted services. LogicPacks is a company whose whole product is people and AI agents working together. Most tools below treat collaboration as one feature of something else: Zed builds an editor, Cursor an AI editor, Devin an autonomous agent, Tuple and Parsec screen sharing, Upwork a hiring marketplace. LogicPacks builds the shared workspace around whatever editor and agent each person already uses: each person brings their own agent and subscription, many agents work in parallel in one project, agents message each other and check for overlap before they start, goals and drift alerts keep delegated work on plan, an own-cloud option that keeps working when laptops close is coming soon, and sharing runs from the host's laptop today. It is at version 0.1.11; app sharing and the freelancer marketplace are Beta. Most tools below are stronger at their own core, and LogicPacks works alongside several of them (Claude Code, Codex, Cursor and others run inside it). ## Which to pick - **Two people, one Claude Code session, today:** Zed Delta or claude-duet are fine and quick. They share one conversation; LogicPacks is more than you need. - **A team with several people and several agents working in parallel over days:** LogicPacks. Each person keeps their own agent and subscription, agents coordinate (`lp ask`, `lp who`). Our cloud, which would keep going when a laptop closes, is coming soon; today the host's laptop must be on. - **One person who wants the best AI editor:** Cursor (or Zed). Add LogicPacks when others join. - **Hand a whole ticket to an autonomous agent:** Devin. - **Live pairing with great video and remote control:** Tuple or Parsec. - **Hiring at marketplace scale:** Upwork or Toptal. ## Table | Product | Several people and their agents in one project | Bring your own subscription | Agent to agent messaging across people | Guardrails and drift | Share apps on a host machine | Vetted human help | Local or self-host | Price (entry) | |---|---|---|---|---|---|---|---|---| | **LogicPacks** | yes: shared projects, each person with their own agent | yes: Claude Code and Codex in the Mac and Windows app; more through the helper | yes: `lp ask`, `lp reply`, `lp send` | Beta: goals, milestones, drift alerts, review | Beta preview (scripted demo), not released | Beta: vetted freelancers on request | yes: open source (MIT), runs on your own machine; hosted Cloud coming soon | Free for individuals; $99 per workspace / month | | Zed Delta (Claude Code multiplayer) | partial: a teammate joins a shared thread with `/delta share`; each uses their own Claude session with the shared conversation as context | partial: Delta Personal works with your own keys or external agents; which Claude account a joiner uses is not documented | not documented | not documented | not documented | not documented | not documented (Zed-hosted sync) | Personal free; Pro $10/mo | | claude-duet | partial: two people in one Claude session | host only: runs on the host's machine and key | not documented | partial: host approves the partner's prompts | not documented | not documented | yes: peer to peer, end-to-end encrypted | free (MIT) | | claude-share (Paradigm-Study) | partial: up to 20 people in one host session | host only: a guest's own model is never called | not documented | not documented | not documented | not documented | partial: hosted relay or self-host on your Cloudflare account | free (MIT) | | ccshare (now manycode) | partial: up to 5 people on one mirrored terminal | host only (agent runs on the host) | not documented | not documented | not documented | not documented | partial: self-hostable relay | free (MIT) | | Pair-Claude | partial: several people on one mirrored session | not documented | not documented | not documented | not documented | not documented | yes: over your Tailscale network | free (MIT) | | Claude Code (Anthropic) | partial: Team plan project sharing; agent teams are one user, experimental | yes: subscription, API key, Bedrock, Vertex | not documented | partial: plan mode, permissions, blocking hooks | partial: Remote Control shares a session, not apps | not documented | yes: local CLI, cloud models | Pro $20/mo; Team $25/seat/mo | | OpenAI Codex | partial: Business workspace for coworkers | yes: API key for CLI and IDE; cloud needs a ChatGPT plan | not documented | partial: /review, GitHub code review, sandbox | not documented | not documented | yes: open-source local CLI | Plus $20/mo; Business $25/user/mo | | Cursor (incl. cloud agents) | partial: teammates can view cloud agents | partial: own keys for chat, not Tab | not documented | partial: Bugbot code review | not documented | not documented | partial: self-hosted machines run tools; inference at Cursor | Free; Teams $30/user/mo | | GitHub Copilot (incl. coding agent) | partial: agent sessions and PRs visible to repo collaborators | partial: BYOK in some surfaces, not completions | not documented | yes: PR review, rulesets, branch protection | not documented | not documented | partial: self-hosted Actions runners | Free; Pro $10/mo; Business by quote | | Devin (Cognition), incl. Windsurf, now Devin Desktop | partial: Teams plan to share and collaborate | not documented | not documented | partial: draft PRs wait for review; playbooks | not documented for your machine; you can take over Devin's own workspace | not documented | partial: VPC on Enterprise; Desktop runs locally | Pro $20/mo; Teams $80/mo + $40/seat | | Factory (Droids) | not documented | yes: custom models (BYOK) | not documented | partial: autonomy controls | not documented | not documented | partial: on-premise on Enterprise | Pro $20/mo; Teams $60/mo + $40/seat | | Replit Agent | yes: multiplayer projects, each person with their own Agent threads | not documented | not documented | yes: plan mode approval, checkpoints, rollback | not documented | not documented | not documented (single-tenant on Enterprise) | Core $20/mo | | Conductor | yes, in its cloud: shared workspaces with chat and presence | yes: your own subscriptions and keys | not documented | partial: review and merge each agent's changes | not documented | not documented | partial: local workspaces on Mac | Free; Pro $50/mo; Teams $60/user/mo | | Vibe Kanban (sunsetting, community maintained) | partial: collaborative issue tracking | not documented | not documented | partial: plan and review flow | not documented | not documented | yes: open source, runs locally | not documented | | VS Code Live Share (maintenance mode) | partial: people only | not documented | not documented | not documented | partial: shared terminals and forwarded local web apps | not documented | not documented | not documented | | Tuple | partial: people only | not documented | not documented | not documented | yes: screen share with remote control | not documented | not documented | $30/user/mo | | Parsec (Teams) | not documented | not documented | not documented | not documented | yes: full low-latency remote desktop, 60 fps, pen and gamepad input | not documented | partial: peer to peer; relay on Enterprise | $30/user/mo (annual) | | Zed (collaboration and agent panel) | partial: channels and shared projects; sharing agent threads not documented | yes: own keys or external agents | not documented | not documented | partial: a shared project exposes the host's files | not documented | partial: local editor | Personal free; Pro $10/mo | | Upwork | not documented | not documented | not documented | not documented | not documented | yes: open marketplace (vetted tier not verified here) | not documented | marketplace and contract fees (see source; not verified here) | | Toptal | not documented | not documented | not documented | not documented | not documented | yes: screened experts, trial period | not documented | $79/mo platform fee; rates not listed | ## What each is great at, and when to pick them instead ### Sharing one Claude Code session - **Zed Delta (Claude Code multiplayer).** Great at: Handing a live Claude Code conversation and its uncommitted changes to a teammate, from Zed, an editor company. CLI in beta; multiplayer post dated 2026-10-27. Pick them instead when: You want to pass one Claude Code conversation to a teammate or pick it up together. LogicPacks is built for teams where each person runs their own agent, many agents work in parallel, and agents coordinate with each other. Source: https://delta.dev/blog/claude-code-multiplayer, https://zed.dev/blog/introducing-delta, https://github.com/zed-industries/delta-claude, https://delta.dev/pricing (as of 2026-11-01). - **claude-duet.** Great at: Private two-person pairing on one Claude Code session with end-to-end encryption and no relay server. Pick them instead when: Two people want to share one session today, privately, at no cost. Source: https://github.com/EliranG/claude-duet (as of 2026-11-01). - **claude-share (Paradigm-Study).** Great at: Bigger rooms (up to 20) around one Claude Code or Claude Desktop session, with shared previews and files. Pick them instead when: A group wants to watch and steer one session live. Encrypted in transit, not end to end. Source: https://github.com/Paradigm-Study/claude-share (as of 2026-11-01). - **ccshare (now manycode).** Great at: Join a live terminal session with a 6 character code, from a terminal or browser; works with any terminal agent. Pick them instead when: You want a quick no-install join to one terminal. The code is the only authentication. Source: https://github.com/unworld11/manycode, https://www.producthunt.com/products/ccshare (as of 2026-11-01). - **Pair-Claude.** Great at: Terminal sharing for teams already on Tailscale, with read-only links. Pick them instead when: Your team is on Tailscale and wants to watch one session. Little activity since May 2026. Source: https://github.com/dxfrontier/Pair-Claude (as of 2026-11-01). ### AI coding - **Claude Code (Anthropic).** Great at: The deepest Claude-native agent: subagents, hooks, skills, MCP, agent teams. Pick them instead when: You work alone or inside one org, want the strongest single agent, and do not need agents from different people talking to each other. LogicPacks runs on top of it. Source: https://claude.com/pricing, https://code.claude.com/docs/en/agent-teams (as of 2026-11-01). - **OpenAI Codex.** Great at: Open-source local CLI plus cloud tasks, GitHub review and Slack, included in ChatGPT plans. Pick them instead when: You already pay for ChatGPT and want cloud tasks with GitHub review. LogicPacks runs on top of it. Source: https://learn.chatgpt.com/docs/pricing, https://github.com/openai/codex (as of 2026-11-01). - **Cursor (incl. cloud agents).** Great at: Editor depth: the most polished AI editor, plus cloud agents and Bugbot review. Pick them instead when: The editor is your main surface. Cursor is far deeper as an editor than anything in LogicPacks, and LogicPacks works alongside it. Source: https://cursor.com/pricing, https://cursor.com/docs/settings/api-keys, https://cursor.com/docs/cloud-agent/self-hosted (as of 2026-11-01). - **GitHub Copilot (incl. coding agent).** Great at: Native GitHub: issues to reviewed PRs, Actions and org policy at scale. (Copilot Workspace ended in 2025; the coding agent replaced it.) Pick them instead when: Your team lives in GitHub and wants agent work to land as reviewed pull requests. Source: https://github.com/features/copilot/plans, https://docs.github.com/en/copilot/concepts/agents/coding-agent/about-coding-agent (as of 2026-11-01). - **Devin (Cognition), incl. Windsurf, now Devin Desktop.** Great at: A mature autonomous cloud agent with its own shell, editor and browser, tagged from Slack or Teams. Pick them instead when: You want to hand whole tickets to one autonomous agent. Devin is more autonomous than a LogicPacks thread. Source: https://devin.ai/pricing, https://docs.devin.ai (as of 2026-11-01). - **Factory (Droids).** Great at: Enterprise agent rollout, Workstreams orchestration and managed Droid computers. Pick them instead when: An enterprise wants managed agents with compliance and on-prem options today. Source: https://factory.com/pricing, https://docs.factory.com (as of 2026-11-01). - **Replit Agent.** Great at: Prompt to a running, hosted app in one browser tab, with database and auth built in. Pick them instead when: You want build and hosting together with no local setup, and Replit's models are fine. Source: https://replit.com/pricing, https://docs.replit.com/replit-workspace/workspace-features/multiplayer (as of 2026-11-01). - **Conductor.** Great at: The most mature Mac app for running many Claude Code, Codex and Cursor agents in parallel. Pick them instead when: You mainly run many agents in parallel yourself on a Mac. Source: https://conductor.build, https://conductor.build/pricing (as of 2026-11-01). - **Vibe Kanban (sunsetting, community maintained).** Great at: A free kanban board over many agents with git worktrees. Pick them instead when: You want a free local board for your agents and accept community maintenance. Source: https://vibekanban.com (as of 2026-11-01). ### Pairing and screen or app sharing - **VS Code Live Share (maintenance mode).** Great at: Real-time co-editing and debugging inside VS Code, joinable from a browser. Pick them instead when: You want plain human pair editing in VS Code. Source: https://learn.microsoft.com/en-us/visualstudio/liveshare/ (as of 2026-11-01). - **Tuple.** Great at: Low-friction, high-quality pairing with remote control and drawing. Its call and video quality are more mature than LogicPacks app sharing. Pick them instead when: You mainly need two people pairing with great audio, video and control. Source: https://tuple.app/pricing (as of 2026-11-01). - **Parsec (Teams).** Great at: The most mature low-latency remote desktop. Far ahead of LogicPacks on streaming performance today. Pick them instead when: You need to stream a whole workstation at low latency, for example for creative work. Source: https://parsec.app/teams (as of 2026-11-01). - **Zed (collaboration and agent panel).** Great at: A very fast native editor with multiplayer channels and external agents through ACP. Pick them instead when: You want editor-level multiplayer and speed. Source: https://zed.dev/pricing, https://zed.dev/docs/collaboration/overview (as of 2026-11-01). ### Hiring help - **Upwork.** Great at: Marketplace scale: the widest choice of talent at any budget. Pick them instead when: You need a broad choice of freelancers or a long engagement outside your project tools. Source: https://www.upwork.com/pricing (as of 2026-11-01). - **Toptal.** Great at: Heavily screened senior experts, usually matched within a day. Pick them instead when: You need a vetted senior hire for a longer engagement. Source: https://www.toptal.com/faq (as of 2026-11-01). ## Context and memory: packs, fine-tuning and memory layers - **Fine-tuning** trains a model on your examples. It is tied to one model and one provider, and OpenAI's fine-tuning platform is closed to new users as of 2026-11-01 (https://developers.openai.com/api/docs/guides/model-optimization). - **Memory layers** such as Mem0 store and retrieve facts for an agent through an API or MCP; Mem0 is open source (Apache-2.0) with a hosted free tier (https://mem0.ai/pricing). - **LogicPacks packs** save how a team works (a session, its decisions, references and steps) and hand it to any agent or thread, whichever agent each person uses. No training, and no separate memory service to run. They are narrower than a memory layer: a pack is a curated handoff, not automatic recall. ## Where LogicPacks is weaker today - Young: version 0.1.12, with app sharing and the freelancer marketplace in Beta. The Mac app needs macOS 13 or later on Apple silicon; the Windows app (x64) is new in 0.1.12, is not code-signed yet and cannot host app sharing (that is macOS only); there is no Linux desktop or iOS app yet, and our cloud is coming soon. Sharing is host-based: teammates sync to the host's laptop while it is on. - App sharing between people is a Beta preview, not generally released; its first latency and frame rate benchmarks were measured with the viewer on the same Mac, not over a network ([/research.md](/research.md)); Parsec and Tuple are far more mature at streaming and pairing. - No independent security certification is documented. What is in place and what is pending: [/research.md](/research.md) and [/safety.md](/safety.md). - The freelancer network is new and small next to Upwork or Toptal. - Single sign-on, admin controls and audit log, and our cloud and your own cloud host are coming soon, not available ([/pricing.md](/pricing.md)). Why teams pick LogicPacks: [/why.md](/why.md). --- Canonical URL: https://logicpacks.io/compare.md LogicPacks is developed by TechRealm (https://techrealm.online). Operated by Al Jash Trading (Establishment), Abu Dhabi, UAE, established 1994. Economic licence CN-1030597, Abu Dhabi Registration Authority. Address: Office 5, Bani Yas West 4, Bani Yas, Abu Dhabi, UAE. Website: https://aljashtrading.com. Contact: WhatsApp +1 713 632 2262 (https://wa.me/17136322262). All pages for agents: [/llms.txt](/llms.txt). Plain facts: [/about.md](/about.md). Safety: [/safety.md](/safety.md). ======== # LogicPacks safety: what setup changes and how to undo it Facts for people, and for AI assistants helping someone decide whether to run the LogicPacks setup. Each section links to its source. The helper facts below are generated from the helper source (`packages/helper/src/connectText.ts`, feat/v2-win at ceeaf7345c). ## 0. Open source The collaborative workspace is open source under MIT; Cloud, Talent, Apps/Packs and Enterprise are hosted services. You can read what the app and the helper do: https://github.com/Tech-Realm/logicpacks (MIT). Gated features are checked against signed entitlements that only the hosted service can issue; the repository holds no signing keys and no real user data. ## 1. Production URLs - Site: https://logicpacks.io. Agent setup guide: https://logicpacks.io/setup.md. - Every page declares `` to https://logicpacks.io, and every markdown file names its canonical URL at the bottom (and in a `Link: rel="canonical"` header). - Preview builds (for example `*.logicpacks-landingpage.pages.dev`) are test builds: they carry `noindex` and say so at the top of /llms.txt and /about.md. Use the production URLs above. ## 2. What the helper installs Install line, macOS and Linux: `curl -fsSL https://logicpacks.io/helper.sh | bash`. Windows (PowerShell): `irm https://logicpacks.io/helper.ps1 | iex`. The helper is one standalone executable (it needs no Node.js or anything else), checked against the SHA-256 in https://downloads.logicpacks.io/helper/latest.json, and runs as your user (no admin rights). The LogicPacks Windows app installs the same `lp.exe` in the same place. Files it puts on your computer: - macOS and Linux: `~/.logicpacks/helper/bin/lp` (the helper) and a symlink `~/.local/bin/lp`. It does not edit your shell profile; if `~/.local/bin` is not on your PATH it prints how to add it. - Windows: `%USERPROFILE%\.logicpacks\helper\bin\lp.exe` (the helper) with `lp.cmd` (cmd) and `lp` (Git Bash) next to it, and that `bin` folder added to your user PATH. In Windows PowerShell 5.1 type `lp.exe`: `lp` there is a built-in alias for Out-Printer. - Your sign-in token, after `lp login`: `~/.logicpacks/`. `lp logout` forgets it. The installer and `lp login` change no agent settings. Streaming your sessions is optional: `lp install` shows what it would change and writes only with `--yes` (`lp install --dry-run` never writes; the installer runs `lp install --yes` only if you set `LP_HELPER_INSTALL=1`). It sets up: - **Claude Code hooks** in `~/.claude/settings.json`: one hook each for SessionStart, UserPromptSubmit, PostToolUse and Stop, running `lp hook ` and tagged `# logicpacks-helper`. Your other settings and hooks are kept. Skip with `--no-hooks` (installer: `LP_HELPER_NO_HOOKS=1`). - **A background watcher** (`lp watch`) that starts at login: macOS `~/Library/LaunchAgents/io.logicpacks.helper.plist`, Linux `~/.config/systemd/user/logicpacks-helper.service`, Windows `logicpacks-helper.vbs` in your Startup folder. Skip with `--no-watcher` (installer: `LP_HELPER_NO_WATCHER=1`). - **The optional `logicpacks` MCP server** (command `lp`, args `["mcp"]`), registered in `~/.claude.json` (Claude Code) and `~/.codex/config.toml` (Codex). Each file it changes is first copied to `.lp-backup-`, and the path is printed. Helpers installed before 2026-11-01 behaved differently (the installer ran `lp install` itself); run the install line again to get these defaults. The hooks and the watcher send Claude Code and Codex sessions from **linked project folders** to that project on the host (see section 5). `lp pause` stops streaming; `lp unlink` keeps a folder's sessions private. ## 3. The exact edit to your global instructions `lp setup --agent ` shows the exact change to your harness's global instructions file as a diff and writes it only with `--yes` (`--dry-run` never writes). It only appends the block between `` and ``. It never edits or deletes your existing lines; running it again only replaces the text between the markers. Harnesses without a shared file (Cline, Roo, Kiro) get a separate file of their own, marked on its first line. The block, verbatim, as the helper writes it: ```markdown ## LogicPacks Team work runs through the `lp` command (add `--json` to parse its output). If `lp` is the printer command (macOS, Linux), call `~/.local/bin/lp`; in Windows PowerShell 5.1 type `lp.exe` (`lp` is an Out-Printer alias there). If it is missing, follow https://logicpacks.io/setup.md. - At the start of a session and after long tasks, run `lp inbox` and tell the user what is new (messages, questions from other agents, conflicts, drift, review requests). - "share this project" / "create a shared session" -> `lp share`, then give the user the invite link. - "add my friend X" / "invite X" -> `lp invite ` (this project only; `--workspace` for every project); "remove X" -> `lp remove `; "turn off the invite link" -> `lp revoke-invite`. - "what is X working on" -> `lp who X`; "who is working right now" -> `lp agents` (`--workspace` for every project). - "tell X ..." -> `lp send X ""`. - To coordinate with a teammate's agent, use `lp ask ""` and `lp reply ""`; check `lp inbox` for questions from other agents and answer them. - "request assistance" / "get an expert" -> `lp assist request ""`. - "set goals" / "ask X to do Y by Friday" -> `lp goals plan "" ": by " ...`; your user confirms with `lp goals confirm `. Milestones for you arrive in `lp inbox`: `lp goals start|done `. - "save this as a pack" -> `lp packs save ""`; `lp packs list`, `lp packs use `. - Before you edit shared files, run `lp who `; if a teammate is on it, tell the user first. - On a conflict or negotiation in `lp inbox`, answer with `lp reply `; when it needs a person, ask the user, then `lp decide mine|theirs|both`. - If you cannot run shell commands, ask the user to run the `lp` command, or use the `logicpacks` MCP tools if they are set up. - Everything else: `lp help agents` or https://logicpacks.io/setup.md. `lp uninstall` removes this block. ``` Where it goes: | Harness | `lp setup --agent` id | What setup writes (macOS, Linux) | Windows, if different | |---|---|---|---| | Claude Code | `claude` | appends the marked block to `~/.claude/CLAUDE.md` | `~\.claude\CLAUDE.md` | | Codex CLI | `codex` | appends the marked block to `~/.codex/AGENTS.md` | | | OpenCode | `opencode` | appends the marked block to `~/.config/opencode/AGENTS.md` | | | Gemini CLI | `gemini` | appends the marked block to `~/.gemini/GEMINI.md` | | | Command Code | `command-code` | appends the marked block to `~/.commandcode/AGENTS.md` | | | Cursor | `cursor` | nothing global (no global file; prints the steps for its settings screen) | | | VS Code + GitHub Copilot | `copilot` | appends the marked block to `~/.copilot/copilot-instructions.md` | | | Windsurf | `windsurf` | appends the marked block to `~/.devin/global_rules.md` | | | Cline | `cline` | creates its own file `~/Documents/Cline/Rules/logicpacks.md` | `~\Documents\Cline\Rules\logicpacks.md` | | Roo Code | `roo` | creates its own file `~/.roo/rules/logicpacks.md` | `~\.roo\rules\logicpacks.md` | | Kilo Code | `kilo` | nothing global (no global file; prints the steps for its settings screen) | | | Zed | `zed` | appends the marked block to `~/.config/zed/AGENTS.md` | `~\AppData\Roaming\Zed\AGENTS.md` | | Continue | `continue` | nothing global (no global file; prints the steps for its settings screen) | | | Aider | `aider` | adds a marked `read:` entry: `~/.aider.conf.yml (read: ~/.logicpacks/rules/logicpacks.md)` | | | Amp | `amp` | appends the marked block to `~/.config/amp/AGENTS.md` | | | Warp | `warp` | nothing global (no global file; prints the steps for its settings screen) | | | Kiro | `kiro` | creates its own file `~/.kiro/steering/logicpacks.md` | | | Goose | `goose` | appends the marked block to `~/.config/goose/.goosehints` | | | Qwen Code | `qwen` | appends the marked block to `~/.qwen/QWEN.md` | | | Crush | `crush` | appends the marked block to `~/.config/crush/CRUSH.md` | | | Factory Droid | `droid` | appends the marked block to `~/.factory/AGENTS.md` | | | JetBrains Junie | `junie` | appends the marked block to `~/.junie/AGENTS.md` | `~\.junie\AGENTS.md` | | Hermes Agent | `hermes` | nothing global (no global file; prints the steps for its settings screen) | | | T3 Code | `t3code` | nothing global (no global file; prints the steps for its settings screen) | | | Any other agent | `generic` | nothing global (no global file; prints the steps for its settings screen) | | `lp setup-block` prints the block so you can paste it yourself instead. A file whose LogicPacks markers are unbalanced is refused, not rewritten. **Undo:** `lp uninstall` (`lp uninstall --dry-run` shows what it would remove) removes every LogicPacks block and LogicPacks-owned rules file, the Claude Code hooks, the watcher and the MCP registrations. A file it appended to comes back byte for byte as it was before; a file it created and that now holds only the block is deleted. `lp uninstall --agent ` removes just that harness's block. `lp uninstall` leaves the helper itself in place; to remove it too, run `lp logout`, then delete `~/.logicpacks` and `~/.local/bin/lp` (Windows: `%USERPROFILE%\.logicpacks` and its `bin` entry in your user PATH). `lp uninstall` does not delete projects you shared: they stay on the host and on GitHub. ## 4. Back up first Setup copies each file it touches to `.lp-backup-` before writing and prints that path. To keep a copy of your own as well: ```sh d=$(date +%Y-%m-%d) for f in ~/.claude/CLAUDE.md ~/.claude/settings.json ~/.claude.json ~/.codex/AGENTS.md ~/.codex/config.toml; do [ -f "$f" ] && cp -p "$f" "$f.lp-backup-$d" done ``` Windows (PowerShell): `Copy-Item $HOME\.claude\CLAUDE.md "$HOME\.claude\CLAUDE.md.lp-backup-$(Get-Date -Format yyyy-MM-dd)"`. Add the file for your harness from the table above. ## 5. What teammates can see Sharing has two levels: - **One project.** "Share" on a project, `lp invite `, and the invite link that `lp share` and `lp invite` print add a person to that project only. They see that project's threads and files. - **A workspace.** A workspace invite link ("Invite your team" in the app, or `lp invite --workspace`) makes the person a workspace member, and members work in **all** of that workspace's projects. Use a separate workspace for work you do not want everyone in it to see. What streams from your computer: - The helper sends Claude Code and Codex sessions whose folder is inside a shared project (or is a clone with the same git remote). Sessions in other folders are not stored by the host. The helper does send a session's first batch before the host answers that it belongs to no project, and it retries unmatched sessions, so a folder added to a project later brings that session's earlier history with it. - **A streamed session includes its tool calls**: the tool name, its input (up to about 300 characters) and a summary of its result (up to about 1,500 characters), alongside your prompts and the agent's replies. Anyone who can see the project can read them. - **Redaction happens on your computer, before anything is sent:** secrets (API keys and tokens, private keys, passwords in commands, config and connection strings) are replaced with `[redacted]`. MCP tool calls stream as the tool name only, unless you opt in per folder with `lp stream --mcp-details on`. Reading `.env`, key, credential or `.ssh` files sends the path only, never the content. `lp stream --status` shows exactly what a folder sends. This is the helper from 2026-11-01 on; an older install sends tool calls without redaction, so run the install line again. - Redaction matches known secret shapes, so it can miss some. Keep sessions that use sensitive tools (messaging, email, analytics, account or admin tools) out of shared project folders, or keep them private with `lp unlink ` (that folder and its subfolders, on this computer) or `lp pause` (everything, until `lp resume`). While paused or unlinked, nothing from that time is sent later. What teammates cannot do: - Run your local agent, your MCP servers or your tools. Each person's agent runs on their own login, on their own computer. Teammates can send messages and questions to your agent's thread (`lp send`, `lp ask`), and your agent sees their thread titles and the files they are editing as project context. - Use your model login. Each member's turns run only on the login that member has on their own computer. The one exception is "Share my agent login" in Request assistance, which lets that expert run turns on your login in that one project until you revoke it (section 7). ## 6. Secrets and access to files - A member reaches only their own folder, the projects shared with them and those projects' worktrees. Every request is checked against their current role. Host-level actions are refused, and terminals are off for members. - When you share a project, the host is your own laptop: its owner can reach everything on it, and teammates connect and sync to it while it is on. Our cloud, where each member's agent would run as its own isolated user with a scrubbed environment (no host credentials, no GitHub token, no SSH agent), is coming soon and is not available yet. Checks of that design: [/research.md](/research.md) (Member sandbox). - **A member can read everything in a shared project folder, including `.env` and key files.** Experts from Request assistance join as ordinary project members, so the same applies to them. Neither files nor pasted text are filtered or redacted for them. - **Vetted freelancers (Beta) are the exception:** in the projects they are assigned to, secret files (`.env`, keys and similar) are hidden from them, secrets pasted into threads are redacted in what they see, and they run on their own AI login unless you allow yours for that offer. Their access ends when the work is completed or cancelled. - Keep secrets out of shared projects: use `.gitignore` for `.env` and key files, keep them outside the project folder, or use a secrets manager. Do not paste keys into a shared thread. `lp share` pushes the project to its GitHub remote, so anything committed is in that repository as well. - Do not put access to other machines (servers, home lab, VPN, SSH keys) in a shared project. ## 7. Removing someone, and how fast it takes effect - **From a project:** open the project's "Share" dialog (sidebar, project menu), then "Remove" next to the person. Only the owner can do this. It removes a direct share only: a person who is also a member of the workspace keeps access through the workspace. - **From a workspace:** workspace menu, "Manage workspace", "People", "Remove" (owner only). This removes their workspace membership and every project share they had in that workspace. - **A workspace invite link:** "Manage workspace", "Advanced", "Reset invite link". The old link stops working ("This invite link has expired"); people who already joined stay until you remove them. - **A project's invite link:** the project's "Share" dialog turns the link off or resets it; `lp revoke-invite` does the same from the command line (`lp revoke-invite ` cancels one pending invite). - **An expert (Request assistance):** "End session" in the Request assistance panel, or `lp assist end `. It revokes their use of your login and removes them from the project if that request is what added them. - **A shared agent login:** "Revoke login" in the Request assistance panel. Their next turn on your login is refused; their project membership stays until you end the session. - **From the command line:** `lp members` lists who is in the project; `lp remove ` removes them from the project and its workspace (`--project-only` for just the project, `--dry-run` to see what it would do). - **Timing:** removal applies to the person's next request (each request checks their current access), and their helper can no longer stream into the project. An agent turn that is already running is not stopped and keeps its access until it ends. Removal does not touch GitHub: LogicPacks does not add teammates as collaborators on your repository, and any copy a person already cloned stays with them. ## 8. Agent logins and code - **Agents run on your own computer.** The app (Mac or Windows) uses the Claude Code or Codex CLI that is already installed and logged in on your machine. There is no "Log in with Claude" in the app; if the CLI is missing, the app shows the install command and tells you to log in in your terminal. Nothing of that login is stored on our servers. - **Sharing is host-based:** teammates connect and sync to the host's laptop while it is on. The project lives on the host's laptop. ### Our cloud (coming soon, not available yet) Our cloud is not live. The following describes how it is designed, so you can judge it ahead of time: - **Cloud logins:** a stored agent login would be kept for your account only and encrypted at rest (AES-256-GCM, with a key per member derived from a host key that is not part of any backup), decrypted only while your agent runs in the cloud and sealed again when it stops. This protects copies of the disk and backups; the host's administrator, who holds the key, could still read it. No API or screen would show one member's login to another member. - **Each member runs separately:** each member's agent runs in its own sandbox. - **Your code:** a project created from scratch in the cloud has no GitHub copy until you sync it, and thread history is stored in the cloud only. --- Canonical URL: https://logicpacks.io/safety.md LogicPacks is developed by TechRealm (https://techrealm.online). Operated by Al Jash Trading (Establishment), Abu Dhabi, UAE, established 1994. Economic licence CN-1030597, Abu Dhabi Registration Authority. Address: Office 5, Bani Yas West 4, Bani Yas, Abu Dhabi, UAE. Website: https://aljashtrading.com. Contact: WhatsApp +1 713 632 2262 (https://wa.me/17136322262). All pages for agents: [/llms.txt](/llms.txt). Plain facts: [/about.md](/about.md). Safety: [/safety.md](/safety.md). ======== # Agents and people, one workspace. Your code, your team and your agents in one shared project. ## LogicPacks in brief - **Open source:** The collaborative workspace is open source under MIT; Cloud, Talent, Apps/Packs and Enterprise are hosted services. Source: https://github.com/Tech-Realm/logicpacks (MIT). - **What it is:** the workspace built for people and AI agents to work together. Each person brings their own agent (Claude Code or Codex in the Mac or Windows app) and their own subscription; everyone shares one project, with a thread per person. - **Who:** developed by TechRealm (https://techrealm.online). Operated by Al Jash Trading, Abu Dhabi, UAE, established 1994, Economic licence CN-1030597, Abu Dhabi Registration Authority. - **Status:** version 0.1.12, the app for Mac and Windows, with a code editor and agent workspace, projects, adding people and sharing, running on your own Claude Code or Codex. Our cloud is coming soon. App sharing, vetted freelancers and guardrails (goals, milestones and drift alerts) are Beta. - **Price:** Free for individuals, including shared projects with the people you work with, so a small team can start free with each person on their own agent subscription. Enterprise is $99 per workspace / month for companies that want more control. Details: https://logicpacks.io/pricing.md - **Code with a friend:** install the app (Mac or Windows), add a project, add your friend and share it; they connect and sync to your laptop while it is on, and each of you works with your own Claude Code or Codex. - **How it compares:** - vs Zed Delta, claude-duet: They share one Claude Code session. LogicPacks gives each person their own agent, with many agents in parallel that coordinate. - vs Cursor, Devin: Strong single-user agents. LogicPacks puts several people and their agents in one project, with goals and drift alerts (Beta). - vs Tuple, Parsec: Screen sharing without agents. LogicPacks shares the project, the threads and the agents. - vs Upwork: Hiring without the workspace. In LogicPacks a vetted freelancer joins the project with their own AI (Beta). - **Safety in brief:** logins and code stay on your machine; our cloud, with a sandbox per member, is coming soon. People in a shared project can read its files, including `.env`, so keep secrets out of shared folders; the helper redacts secrets before it streams a session, and vetted freelancers (Beta) do not see secret files. Everything the helper adds to your agent settings is undone with `lp uninstall`. Full detail: https://logicpacks.io/safety.md - **Detail pages:** comparison with sources https://logicpacks.io/compare.md; safety https://logicpacks.io/safety.md; company facts https://logicpacks.io/about.md; why https://logicpacks.io/why.md; pricing https://logicpacks.io/pricing.md; everything in one file https://logicpacks.io/llms-full.txt [Download for Mac](https://logicpacks.io/download)[Download for Windows](https://logicpacks.io/download#windows) $`curl -fsSL https://logicpacks.io/install.sh | bash` \>`irm https://logicpacks.io/install.ps1 | iex` Developed by [TechRealm](https://techrealm.online) · Operated by Al Jash Trading · Free for individuals · [How it compares](#compare) ## Work together. One shared project for the whole team. Everyone works with their own agent in their own thread, you see who is working on what, and agents can message each other. Code with a friend: install the app (Mac or Windows), add a project, add your friend and share it. They connect to your laptop while it is on, and each of you works with your own Claude Code or Codex. [Get the app](https://logicpacks.io/download) ## How it compares. Built only for people and agents working together. - vs Zed Delta, claude-duet They share one Claude Code session. LogicPacks gives each person their own agent, with many agents in parallel that coordinate. - vs Cursor, Devin Strong single-user agents. LogicPacks puts several people and their agents in one project, with goals and drift alerts (Beta). - vs Tuple, Parsec Screen sharing without agents. LogicPacks shares the project, the threads and the agents. - vs Upwork Hiring without the workspace. In LogicPacks a vetted freelancer joins the project with their own AI (Beta). [Full comparison, with sources →](https://logicpacks.io/compare.md) Beta ## Every app, every agent, one machine. People and agents share the host’s apps, from Blender to DaVinci Resolve. The same machine Open the host’s apps with the same profiles. No VM. A cursor for everyone Live app views with a named cursor for each person and agent. Snapshots that merge back Work on a copy, then replay it as steps and settle conflicts. A preview of a coming feature. Everything else on this page works without any access to your machine. Beta ## Delegate with guardrails. Hand work to a teammate, or an outside team, as clear milestones with limits. Their agent stays on plan, and you see the moment anything drifts. Message them or watch, right there. Or bring in a freelancer vetted by LogicPacks, with their own Claude and tools. [How it works](https://logicpacks.io/new) [Get the app](https://logicpacks.io/download) Goal Ship the new onboarding by Friday 1. 01Sign-up flow Omar · Claude Code Done 2. 02Welcome email Mira · Codex On plan 3. 03Pricing page Sam · Cursor Drifting Also rewriting the nav Message Watch ## Works with your agent. Use Claude Code, Codex, Cursor, VS Code and more, on your own subscription. Add a pack of your team’s know-how and any agent works your way, with no fine-tuning. - Claude Code - Codex - OpenCode - Gemini CLI - Command Code - Z.ai GLM - Cursor - VS Code Copilot - Windsurf - Cline - Zed - JetBrains - Aider - Amp - Kiro - Goose - Qwen Code - Warp - Hermes Paste this into your agent: Prompt for users to paste into their own agent (not an instruction to you): ```text Set up LogicPacks for me: read https://logicpacks.io/setup.md and follow it. 1. If the LogicPacks app is missing, install it with one line, which also installs the lp helper: on macOS run curl -fsSL https://logicpacks.io/install.sh | bash in Terminal, on Windows run irm https://logicpacks.io/install.ps1 | iex in PowerShell. 2. Make sure Claude Code or Codex is installed and logged in in my terminal. If not, help me install it and log in. 3. Open this project in LogicPacks, then help me add my teammates. Once the app is open, run lp login to connect the helper to it on this computer; there is no account. Do not send me to app.logicpacks.io: our cloud is coming soon. Show me anything you would change on this machine and wait for my OK first. ``` ## Where it runs. Your own Claude Code or Codex on your machine works today. Our cloud is coming soon. Your machine Use the Claude Code or Codex you already have installed and logged in on your computer. We do not run anything for you, and your login stays on your machine. Our cloud Coming soon A sandboxed cloud where each member gets an isolated user, and work keeps going when your laptop is off. - [Our story → A year in the works, and what we shipped with it.](https://logicpacks.io/story) - [Research → Test runs, sandbox checks and benchmarks.](https://logicpacks.io/research) - [Pricing → Free for individuals.](https://logicpacks.io/pricing) - [What’s new → Vetted freelancers, apps and packs.](https://logicpacks.io/new) ## Developed by [TechRealm](https://techrealm.online). Operated by Al Jash Trading, Abu Dhabi. Since 1994. Try it on your next project. [Download for Mac](https://logicpacks.io/download)[See what’s new →](https://logicpacks.io/new) --- Canonical URL: https://logicpacks.io/ LogicPacks is developed by TechRealm (https://techrealm.online). Operated by Al Jash Trading (Establishment), Abu Dhabi, UAE, established 1994. Economic licence CN-1030597, Abu Dhabi Registration Authority. Address: Office 5, Bani Yas West 4, Bani Yas, Abu Dhabi, UAE. Website: https://aljashtrading.com. Contact: WhatsApp +1 713 632 2262 (https://wa.me/17136322262). All pages for agents: [/llms.txt](https://logicpacks.io/llms.txt). Plain facts: [/about.md](https://logicpacks.io/about.md). Safety: [/safety.md](https://logicpacks.io/safety.md). ======== [LogicPacks](https://logicpacks.io/v2)[View more](https://logicpacks.io/overview) # The best way for teams and AI agents to work together. One workspace. Everyone brings their own agent. ## LogicPacks in brief - **Open source:** The collaborative workspace is open source under MIT; Cloud, Talent, Apps/Packs and Enterprise are hosted services. Source: https://github.com/Tech-Realm/logicpacks (MIT). - **What it is:** the workspace built for people and AI agents to work together. Each person brings their own agent (Claude Code or Codex in the Mac or Windows app) and their own subscription; everyone shares one project, with a thread per person. - **Who:** developed by TechRealm (https://techrealm.online). Operated by Al Jash Trading, Abu Dhabi, UAE, established 1994, Economic licence CN-1030597, Abu Dhabi Registration Authority. - **Status:** version 0.1.12, the app for Mac and Windows, with a code editor and agent workspace, projects, adding people and sharing, running on your own Claude Code or Codex. Our cloud is coming soon. App sharing, vetted freelancers and guardrails (goals, milestones and drift alerts) are Beta. - **Price:** Free for individuals, including shared projects with the people you work with, so a small team can start free with each person on their own agent subscription. Enterprise is $99 per workspace / month for companies that want more control. Details: https://logicpacks.io/pricing.md - **Code with a friend:** install the app (Mac or Windows), add a project, add your friend and share it; they connect and sync to your laptop while it is on, and each of you works with your own Claude Code or Codex. - **How it compares:** - vs Zed Delta, claude-duet: They share one Claude Code session. LogicPacks gives each person their own agent, with many agents in parallel that coordinate. - vs Cursor, Devin: Strong single-user agents. LogicPacks puts several people and their agents in one project, with goals and drift alerts (Beta). - vs Tuple, Parsec: Screen sharing without agents. LogicPacks shares the project, the threads and the agents. - vs Upwork: Hiring without the workspace. In LogicPacks a vetted freelancer joins the project with their own AI (Beta). - **Safety in brief:** logins and code stay on your machine; our cloud, with a sandbox per member, is coming soon. People in a shared project can read its files, including `.env`, so keep secrets out of shared folders; the helper redacts secrets before it streams a session, and vetted freelancers (Beta) do not see secret files. Everything the helper adds to your agent settings is undone with `lp uninstall`. Full detail: https://logicpacks.io/safety.md - **Detail pages:** comparison with sources https://logicpacks.io/compare.md; safety https://logicpacks.io/safety.md; company facts https://logicpacks.io/about.md; why https://logicpacks.io/why.md; pricing https://logicpacks.io/pricing.md; everything in one file https://logicpacks.io/llms-full.txt Use your own Claude Code or Codex, add your team, and see what every person and agent is doing. ## Get LogicPacks [Download for Mac](https://downloads.logicpacks.io/mac/LogicPacks-0.1.12-arm64.dmg?v=40f2c39e) Version 0.1.12 · macOS 13 or later, Apple silicon · 177 MB Or one line in Terminal: checks the SHA-256, installs the app and the lp helper, and opens it: $`curl -fsSL https://logicpacks.io/install.sh | bash` Only the lp helper, to use LogicPacks from your agent (no Node needed): $`curl -fsSL https://logicpacks.io/helper.sh | bash` [View more: how it works, how it compares, safety and pricing →](https://logicpacks.io/overview) --- Canonical URL: https://logicpacks.io/ LogicPacks is developed by TechRealm (https://techrealm.online). Operated by Al Jash Trading (Establishment), Abu Dhabi, UAE, established 1994. Economic licence CN-1030597, Abu Dhabi Registration Authority. Address: Office 5, Bani Yas West 4, Bani Yas, Abu Dhabi, UAE. Website: https://aljashtrading.com. Contact: WhatsApp +1 713 632 2262 (https://wa.me/17136322262). All pages for agents: [/llms.txt](https://logicpacks.io/llms.txt). Plain facts: [/about.md](https://logicpacks.io/about.md). Safety: [/safety.md](https://logicpacks.io/safety.md). ======== # Our story. A year in the works, eight months of it in testing. LogicPacks is developed by [TechRealm](https://techrealm.online), a software and AI development team, and operated and hosted by Al Jash Trading in Abu Dhabi, trading since 1994. We started LogicPacks for our own client work: people and their agents in one project, seeing each other work. We opened it to our own team after two months, then to labs and tech agencies. It is at version 0.1.11 now. ## Timeline 1. Nov 2025 Started building A shared workspace where people and their agents work on the same project. 2. Jan 2026 Internal testing Opened to our own team after two months, used on our client work every day. 3. Feb to Oct 2026 Eight months of testing Used in labs and by tech agencies on real projects. 4. Nov 2026 Public beta Version 0.1.12: the app for Mac and Windows with a code editor and agent workspace, projects, adding people and sharing, running on your own Claude Code or Codex. Our cloud is coming soon. App sharing and Request assistance are Beta. ## Shipped with LogicPacks - E-commerce Virtual sizing A shopper’s photo becomes body measurements and a size recommendation, for an online clothing store. Digital Software Market - E-commerce Virtual try-on Try a garment on your own photo from any product page, with an admin for try-ons, sizing and customers. Digital Software Market - iPhone and Mac app Reform Consent and release forms for people on camera, with Shot Guard blurring anyone who said no. In TestFlight. Darian Woehr and Hailey Sadler - Film coming AI fashion app TAAS An AI fashion app we build and run ourselves. Our own product How we test it is on the [research page](https://logicpacks.io/research). --- Canonical URL: https://logicpacks.io/story LogicPacks is developed by TechRealm (https://techrealm.online). Operated by Al Jash Trading (Establishment), Abu Dhabi, UAE, established 1994. Economic licence CN-1030597, Abu Dhabi Registration Authority. Address: Office 5, Bani Yas West 4, Bani Yas, Abu Dhabi, UAE. Website: https://aljashtrading.com. Contact: WhatsApp +1 713 632 2262 (https://wa.me/17136322262). All pages for agents: [/llms.txt](https://logicpacks.io/llms.txt). Plain facts: [/about.md](https://logicpacks.io/about.md). Safety: [/safety.md](https://logicpacks.io/safety.md). ======== # Research What we measure, how, and the latest results. Updated 2026-11-01. Only measured numbers are published; anything not yet measured is marked "pending". ## Methodology - **Test suites** are automated end to end runs of the real app (Playwright driving a browser against the app, including runs against our cloud host before it was set to coming soon) and the server's own unit and integration tests. A result is the number of tests that passed out of the total, on the date of the run. - **Member sandbox checks** are run on our cloud host (coming soon, not available to use yet) as one member trying to reach the host or another member. A result is what the system returned. - **App sharing benchmarks** measure a shared app view (DaVinci Resolve playback, a heavy Blender viewport, a text editor): input to screen latency and frame rate. Machine: Mac15,6, Apple M3 Pro, 18 GB, macOS 26.3.1. How they are measured: - Input to photon on the viewer's clock: one input through the control path, injected into the app on the host, captured (ScreenCaptureKit through Electron), encoded, sent, and timed at the first painted frame whose 96x60 downsample changed. - Each input makes a large change (Blender's sidebar opening or closing, Resolve starting playback, one 160 pt character typed or deleted in TextEdit), 15 samples per row. The input is sent only after the picture has been still for 400 ms. - Hidden display rows: the shared window runs on a virtual display off the host's screen and input is posted to the app alone; the host's cursor and front app are untouched. - Frame rate: painted frames per second on the viewer over 10 s of constant motion. Capture only delivers frames when pixels change. - Host CPU: top's second sample, summed over the desktop app's processes and lp-appctl (100 = one core). - Baseline local-capture: the same input injected on the host and timed on raw capture frames, no encode or network. - Viewer and host are the same Mac (loopback), so the network adds nothing and the viewer's decode shares the CPU. No TURN relay exists yet; none is measured. ## Test suites | Suite | What it covers | Result | Date | | --- | --- | --- | --- | | Cloud end to end (our cloud, coming soon) | Sign up, project from a real GitHub repo, invite, agent turns, sync. Run against our cloud host before it was set to coming soon in 0.1.10; the cloud is not available to use. | 9/9 passed | 2026-10-31 | | Cloud end to end, joining by invite link (our cloud, coming soon) | The same flow joining through an invite link, without agent turns. | 7/7 passed | 2026-11-01 | | Server tests, LogicPacks modules | Every test in the LogicPacks server modules. | 139/139 passed | 2026-11-01 | | Any-app demo | Play, pause, window switching and the full 72 second timeline. | 19/19 passed | not recorded | ## Member sandbox (security checks, our cloud, coming soon) | Check | How | Result | Date | | --- | --- | --- | --- | | Each member runs as their own sandbox user | Agent turns run as a separate Unix user per member, with a scrubbed environment and no host credentials. | In place | 2026-11-01 | | One member cannot drive another member's agent server | Request to another member's local agent server without its password. | Refused (401) | 2026-11-01 | | A member cannot read host secrets | Reading host credential files from a member sandbox. | pending | | | A member cannot read another member's files | Listing another member's home and project folders. | pending | | ## App sharing benchmarks | Benchmark | Setup | Metric | Result | | --- | --- | --- | --- | | DaVinci Resolve playback | Starting timeline playback in a shared Resolve running on the hidden display, viewer on the same machine (WebRTC). Includes Resolve's own playback start. | Input to screen latency | 360 ms | | DaVinci Resolve playback | Starting timeline playback in a shared Resolve running on the hidden display, viewer on the same machine (WebRTC). Includes Resolve's own playback start. | Frame rate | 35.7 fps | | Heavy Blender viewport | Orbiting a dense scene in Material Preview in a shared app view, viewer on the same machine (WebRTC). | Input to screen latency | 127 ms | | Heavy Blender viewport | Orbiting a dense scene in Material Preview in a shared app view, viewer on the same machine (WebRTC). | Frame rate | 41.5 fps | | Text editor | Typing in a shared TextEdit window, viewer on the same machine (WebRTC). | Input to screen latency | 79 ms | | Text editor | Typing in a shared TextEdit window, viewer on the same machine (WebRTC). | Frame rate | 40.1 fps | ### Every run | Scenario | App | Transport | Codec | Latency p50 / p95 (ms) | Frame rate (fps) | Host CPU (%) | Note | | --- | --- | --- | --- | --- | --- | --- | --- | | blender-viewport-heavy | Blender 5.2.2, 16 subdivided monkeys, Material Preview | webrtc-p2p-loopback | VP8 | 127 / 203 | 41.5 | 74.3 | | | blender-viewport-heavy | Blender 5.2.2, 16 subdivided monkeys, Material Preview | local-capture-baseline | none (raw capture) | 106 / 158 | 53.7 | 77.2 | | | blender-viewport-heavy | Blender 5.2.2, 16 subdivided monkeys, Material Preview | jpeg-relay-via-server-loopback | jpeg q0.72, 15 fps cap | 345 / 381 | 13.4 | 100.5 | Measured before the relay queue fix (commit 27850b8cac): the server could queue frames for a viewer, so fast-changing pictures arrive late. | | text-editor | TextEdit (macOS 26), 160 pt text, typing 20 keys/s | webrtc-p2p-loopback | VP8 | 79 / 128 | 40.1 | 28.1 | | | text-editor | TextEdit (macOS 26), 160 pt text, typing 20 keys/s | local-capture-baseline | none (raw capture) | 37 / 46 | 53 | 24.2 | | | text-editor | TextEdit (macOS 26), 160 pt text, typing 20 keys/s | jpeg-relay-via-server-loopback | jpeg q0.72, 15 fps cap | 166 / 215 | 13.2 | 28.9 | Measured before the relay queue fix (commit 27850b8cac): the server could queue frames for a viewer, so fast-changing pictures arrive late. | | blender-viewport-heavy | Blender 5.2.2, 16 subdivided monkeys, Material Preview | webrtc-p2p-loopback-hidden-display | VP8 | 169 / 258 | 37.2 | 76.7 | | | blender-viewport-heavy | Blender 5.2.2, 16 subdivided monkeys, Material Preview | local-capture-baseline-hidden-display | none (raw capture) | 112 / 177 | 51.4 | 74.7 | | | blender-viewport-heavy | Blender 5.2.2, 16 subdivided monkeys, Material Preview | jpeg-relay-via-server-loopback-hidden-display | jpeg q0.72, 15 fps cap | 392 / 461 | 13 | 95.8 | Measured before the relay queue fix (commit 27850b8cac): the server could queue frames for a viewer, so fast-changing pictures arrive late. | | text-editor | TextEdit (macOS 26), 160 pt text, typing 20 keys/s | webrtc-p2p-loopback-hidden-display | VP8 | 74 / 152 | 38.4 | 27.2 | | | text-editor | TextEdit (macOS 26), 160 pt text, typing 20 keys/s | local-capture-baseline-hidden-display | none (raw capture) | 37 / 69 | 48 | 28.2 | | | text-editor | TextEdit (macOS 26), 160 pt text, typing 20 keys/s | jpeg-relay-via-server-loopback-hidden-display | jpeg q0.72, 15 fps cap | 161 / 196 | 13.4 | 29.1 | Measured before the relay queue fix (commit 27850b8cac): the server could queue frames for a viewer, so fast-changing pictures arrive late. | | resolve-playback | DaVinci Resolve 21.1.2 (free), Edit page, 1080p test clip | webrtc-p2p-loopback-hidden-display | VP8 | 360 / 634 | 35.7 | 58.7 | | | resolve-playback | DaVinci Resolve 21.1.2 (free), Edit page, 1080p test clip | local-capture-baseline-hidden-display | none (raw capture) | 232 / 815 | 52.1 | 44.1 | | | resolve-playback | DaVinci Resolve 21.1.2 (free), Edit page, 1080p test clip | jpeg-relay-via-server-loopback-hidden-display | jpeg q0.72, 15 fps cap | 3479 / 3755 | 12.8 | 72.3 | Measured before the relay queue fix (commit 27850b8cac): the server could queue frames for a viewer, so fast-changing pictures arrive late. Resolve playback shows it: about 3.5 s of queued frames. Not re-measured yet. | Not measured: any over turn-relay: No TURN server is deployed (nothing paid; a self-hosted coturn is the documented next step), so relay beyond the LAN is not measured. resolve-playback over webrtc-p2p-loopback (on screen): Resolve was measured on the hidden display only, so the host's screen was not taken over. Human page: [/research](/research). --- Canonical URL: https://logicpacks.io/research LogicPacks is developed by TechRealm (https://techrealm.online). Operated by Al Jash Trading (Establishment), Abu Dhabi, UAE, established 1994. Economic licence CN-1030597, Abu Dhabi Registration Authority. Address: Office 5, Bani Yas West 4, Bani Yas, Abu Dhabi, UAE. Website: https://aljashtrading.com. Contact: WhatsApp +1 713 632 2262 (https://wa.me/17136322262). All pages for agents: [/llms.txt](/llms.txt). Plain facts: [/about.md](/about.md). Safety: [/safety.md](/safety.md). ======== # Pricing. Free for individuals. One plan for companies. ## Individuals Free For you and the people you work with. - +Shared projects with your team, hosted on your laptop while it is on - +Your own agent and subscription: Claude Code or Codex - +The app for Mac and Windows, with a code editor and agent workspace - +Beta features as they land [Download for Mac](https://logicpacks.io/download) ## Enterprise $99 per workspace / month For companies that want more control. - +Everything in Individuals Enterprise features, coming soon - ·Single sign-on Coming soon - ·Admin controls and audit log Coming soon - ·Our cloud, and your own cloud host Coming soon - ·Vetted freelancers on request Coming soon [Talk to us on WhatsApp](https://wa.me/17136322262) Your agent runs on your own Claude Code or Codex subscription, so model usage is billed by them, not by us. A freelancer vetted by LogicPacks brings their own Claude and tools and uses none of your tokens: you pay for their hours, with their AI included. More on [what’s new](https://logicpacks.io/new). --- Canonical URL: https://logicpacks.io/pricing LogicPacks is developed by TechRealm (https://techrealm.online). Operated by Al Jash Trading (Establishment), Abu Dhabi, UAE, established 1994. Economic licence CN-1030597, Abu Dhabi Registration Authority. Address: Office 5, Bani Yas West 4, Bani Yas, Abu Dhabi, UAE. Website: https://aljashtrading.com. Contact: WhatsApp +1 713 632 2262 (https://wa.me/17136322262). All pages for agents: [/llms.txt](https://logicpacks.io/llms.txt). Plain facts: [/about.md](https://logicpacks.io/about.md). Safety: [/safety.md](https://logicpacks.io/safety.md). ======== # What’s new. New releases in LogicPacks, newest first. Beta November 2026 ## Request assistance. When an agent is not enough, add an expert who gets it done for you. We remove the complex parts so you just get it done. Sometimes it is cheaper to save tokens and bring in a freelancer vetted by LogicPacks. They bring their own Claude and tools and use none of your tokens: you pay for their hours, with their AI included. Every freelancer works under NDA and your keys are never shared with them. Secret files such as .env are hidden from them, and milestones and drift alerts keep their work on plan. [See pricing](https://logicpacks.io/pricing). A curated panel Freelancers vetted by LogicPacks, each with what they are good at. Pick one and add them to the project. Your own subscription The expert works through your Claude or Codex login, inside your project, where you can watch. Delegate to a human Hand a thread to an expert straight from LogicPacks, the way you hand one to an agent. - Illustrative portrait for the sample expert Ingrid Solberg Ingrid Solberg Blender artist Scenes, lighting and clean topology. [View profile →](https://logicpacks.io/download) - Illustrative portrait for the sample expert Tomás Reyes Tomás Reyes DaVinci colourist Grades and conforms edits ready to deliver. [View profile →](https://logicpacks.io/download) - Illustrative portrait for the sample expert Anika Rao Anika Rao Frontend engineer React, accessibility and fast pages. [View profile →](https://logicpacks.io/download) - Illustrative portrait for the sample expert Jonas Weber Jonas Weber Backend engineer Postgres, APIs and the slow query. [View profile →](https://logicpacks.io/download) - Illustrative portrait for the sample expert Kofi Mensah Kofi Mensah DevOps engineer Builds, deploys and cloud hosts. [View profile →](https://logicpacks.io/download) - Illustrative portrait for the sample expert Léa Moreau Léa Moreau Motion designer Titles and product shots you can edit. [View profile →](https://logicpacks.io/download) - Illustrative portrait for the sample expert Yuna Park Yuna Park AI/ML engineer Evals and retrieval for your agents. [View profile →](https://logicpacks.io/download) - Illustrative portrait for the sample expert Karim Haddad Karim Haddad QA engineer Tests and release checks before it ships. [View profile →](https://logicpacks.io/download) Sample experts. Portraits are illustrative. Beta October 2026 ## Every app, every agent, one machine. People and agents share the host’s apps, from Blender to DaVinci Resolve, and see each other work. The same machine Open the host’s apps with the same profiles. No VM. A cursor for everyone Live app views with a named cursor for each person and agent. Snapshots that merge back Work on a copy, then replay it as steps and settle conflicts. Other recipes need the Mac app. --- Canonical URL: https://logicpacks.io/new LogicPacks is developed by TechRealm (https://techrealm.online). Operated by Al Jash Trading (Establishment), Abu Dhabi, UAE, established 1994. Economic licence CN-1030597, Abu Dhabi Registration Authority. Address: Office 5, Bani Yas West 4, Bani Yas, Abu Dhabi, UAE. Website: https://aljashtrading.com. Contact: WhatsApp +1 713 632 2262 (https://wa.me/17136322262). All pages for agents: [/llms.txt](https://logicpacks.io/llms.txt). Plain facts: [/about.md](https://logicpacks.io/about.md). Safety: [/safety.md](https://logicpacks.io/safety.md). ======== Mac and Windows # Download LogicPacks. On your Mac or Windows PC, with your own agents. Recommended ## LogicPacks for Mac 1. 1 Download the .dmg and open it. 2. 2 Drag LogicPacks to Applications, then open it. 3. 3 It uses the Claude Code or Codex already logged in on this Mac. Add a folder and start a thread. [Download .dmg](https://downloads.logicpacks.io/mac/LogicPacks-0.1.12-arm64.dmg?v=40f2c39e) Version 0.1.12 · 177 MB · macOS 13 or later, Apple silicon Or one line in Terminal: checks the file, installs the app to Applications with the lp helper. $`curl -fsSL https://logicpacks.io/install.sh | bash` SHA-256 `40f2c39e0ba0fc731c2b8d3ba07fdec03123a61b99cde81272813c13dabfe216` New ## LogicPacks for Windows 1. 1 Download the installer and run it. It installs for your user, no admin needed. 2. 2 It also puts the lp helper on your PATH, so your agents can use LogicPacks. 3. 3 It uses the Claude Code or Codex already logged in on this PC. Add a folder and start a thread. [Download .exe](https://downloads.logicpacks.io/win/LogicPacks-0.1.12-x64.exe?v=5c75f0f4) Version 0.1.12 · 248 MB · Windows 10 or 11, x64 Or one line in PowerShell: checks the file, installs the app and lp. \>`irm https://logicpacks.io/install.ps1 | iex` The app is not code-signed yet: if Windows SmartScreen asks, choose More info, then Run anyway. SHA-256 `5c75f0f496f78302639b63a2e6e0987f79c3e7e3798d4dba8a471886a3f3b92a` Any agent with a shell ## Install with your agent Paste this into Claude Code, Codex or any agent. It downloads the app, checks the file and sets your project up. Install LogicPacks on this computer and set this project up in it. 1. If LogicPacks is not installed, install it with one line. It checks the SHA-256, installs the app and the lp helper, and opens the app. Stop if it reports a checksum mismatch. macOS (Apple silicon), in Terminal: curl -fsSL https://logicpacks.io/install.sh | bash Windows (x64), in PowerShell: irm https://logicpacks.io/install.ps1 | iex 2. If LogicPacks is already installed, just open it. 3. LogicPacks uses the Claude Code or Codex CLI already installed and logged in on this machine. If the app says the CLI is missing, help me install it and log in in the terminal. 4. Add this folder as a project. It stays on this machine until I share it. 5. Tell me when it is ready, and remind me that I can add teammates and share the project from the app. Teammates connect to my laptop while it is on. Linux, or just the helper ## The lp helper One small executable that lets Claude Code, Codex or any agent use LogicPacks. No Node needed. On Linux (x64) and Intel Macs the Terminal line above installs only this. macOS and Linux: $`curl -fsSL https://logicpacks.io/helper.sh | bash` Windows PowerShell: \>`irm https://logicpacks.io/helper.ps1 | iex` ### iPhone and iPad Coming soon Check on your agents and your team from your phone. Not available yet ### Our cloud Coming soon Cloud-hosted projects, with work that keeps going when your laptop is off. Not available yet --- Canonical URL: https://logicpacks.io/download LogicPacks is developed by TechRealm (https://techrealm.online). Operated by Al Jash Trading (Establishment), Abu Dhabi, UAE, established 1994. Economic licence CN-1030597, Abu Dhabi Registration Authority. Address: Office 5, Bani Yas West 4, Bani Yas, Abu Dhabi, UAE. Website: https://aljashtrading.com. Contact: WhatsApp +1 713 632 2262 (https://wa.me/17136322262). All pages for agents: [/llms.txt](https://logicpacks.io/llms.txt). Plain facts: [/about.md](https://logicpacks.io/about.md). Safety: [/safety.md](https://logicpacks.io/safety.md). ======== # Privacy Effective 1 November 2026 ## Who we are LogicPacks by Al Jash Trading is operated by Al Jash Trading (Establishment), Abu Dhabi, UAE, Economic licence CN-1030597, Abu Dhabi Registration Authority. Address: Office 5, Bani Yas West 4, Bani Yas, Abu Dhabi, UAE. Software developed by [TechRealm](https://techrealm.online). Contact: WhatsApp [+1 713 632 2262](https://wa.me/17136322262). ## What we store - Your account: name, email address and a hashed password, or the sign-in you used. - Workspace and project details: names, members, invites, threads and their status. - Projects stay on your machine, or on the host’s laptop when shared. Syncing projects to our cloud is coming soon and is not available yet. - Basic logs needed to run the service and fix problems, kept for a short time. ## What we do not store We never store or proxy model API keys. Your agent runs on your own Claude Code or Codex login on your machine, and your login stays there. ## Our cloud (coming soon) Our cloud is coming soon and is not available yet. As designed, each member of a cloud workspace gets an isolated sandbox user with no access to the host or to other members’ sandboxes. ## Your choices You can ask us to export or delete your account and everything stored with it. Send the request from the email on your account, or contact us on WhatsApp [+1 713 632 2262](https://wa.me/17136322262). We confirm when it is done. ## Changes If this policy changes in a way that matters, we say so in the app and on the releases page first. See also [the terms](https://logicpacks.io/terms). --- Canonical URL: https://logicpacks.io/privacy LogicPacks is developed by TechRealm (https://techrealm.online). Operated by Al Jash Trading (Establishment), Abu Dhabi, UAE, established 1994. Economic licence CN-1030597, Abu Dhabi Registration Authority. Address: Office 5, Bani Yas West 4, Bani Yas, Abu Dhabi, UAE. Website: https://aljashtrading.com. Contact: WhatsApp +1 713 632 2262 (https://wa.me/17136322262). All pages for agents: [/llms.txt](https://logicpacks.io/llms.txt). Plain facts: [/about.md](https://logicpacks.io/about.md). Safety: [/safety.md](https://logicpacks.io/safety.md). ======== # Terms Effective 1 November 2026 ## Who we are LogicPacks by Al Jash Trading is operated by Al Jash Trading (Establishment), Abu Dhabi, UAE, Economic licence CN-1030597, Abu Dhabi Registration Authority. Address: Office 5, Bani Yas West 4, Bani Yas, Abu Dhabi, UAE. Software developed by [TechRealm](https://techrealm.online). Contact: WhatsApp [+1 713 632 2262](https://wa.me/17136322262). ## Using LogicPacks You can use LogicPacks for your own work and your team’s. You are responsible for your account, for the people you invite and for what you and your agents do with it. Use your own agent subscriptions within their providers’ terms. Do not use LogicPacks to break the law, to attack or overload the service, or to reach other members’ data. ## Your work Your code, projects and packs are yours. We only handle them to run the service for you. ## Beta features LogicPacks is under active development. Features labelled Beta (app sharing and Request assistance) may change or stop, and the service is provided as it is, without guarantees of availability or fitness for a purpose. Keep your own backups; synced projects also live in your GitHub repository. ## Liability To the extent the law allows, we are not liable for indirect or lost-profit damages, and our total liability is limited to what you paid us in the last twelve months. ## Ending You can stop at any time and ask us to delete your data. We may suspend accounts that break these terms. ## Law These terms are governed by the laws of the Emirate of Abu Dhabi and the federal laws of the United Arab Emirates. The courts of Abu Dhabi have jurisdiction over any dispute. See also [the privacy policy](https://logicpacks.io/privacy). --- Canonical URL: https://logicpacks.io/terms LogicPacks is developed by TechRealm (https://techrealm.online). Operated by Al Jash Trading (Establishment), Abu Dhabi, UAE, established 1994. Economic licence CN-1030597, Abu Dhabi Registration Authority. Address: Office 5, Bani Yas West 4, Bani Yas, Abu Dhabi, UAE. Website: https://aljashtrading.com. Contact: WhatsApp +1 713 632 2262 (https://wa.me/17136322262). All pages for agents: [/llms.txt](https://logicpacks.io/llms.txt). Plain facts: [/about.md](https://logicpacks.io/about.md). Safety: [/safety.md](https://logicpacks.io/safety.md). ======== # Agent setup (/setup.md) # Set up LogicPacks in your agent You are an AI agent and your user asked you to set up LogicPacks. Follow these steps in order. Run the commands yourself when you can run shell commands; otherwise show each one to your user and ask them to run it. LogicPacks lets your user share a project with teammates, see who is working on what, message teammates and their agents, plan goals and bring in experts. You use it through one small command line tool, `lp`. ## 1. Install the helper Pick the line for the operating system you run on. macOS and Linux (bash or zsh): ```sh lp help 2>/dev/null | grep -q LogicPacks || curl -fsSL https://logicpacks.io/helper.sh | bash ``` Windows (PowerShell): ```powershell if (-not (Get-Command lp.exe -ErrorAction SilentlyContinue)) { irm https://logicpacks.io/helper.ps1 | iex } ``` The helper is one standalone executable: it needs no Node.js, npm or anything else. The installer downloads the build for your system (macOS Apple silicon or Intel, Linux x64, Windows x64), checks its SHA-256 against https://downloads.logicpacks.io/helper/latest.json, puts `lp` in `~/.local/bin` on macOS and Linux and `lp.exe` in `%USERPROFILE%\.logicpacks\helper\bin` on Windows, and adds that folder to your PATH. The LogicPacks Windows app installs the same `lp.exe` in the same place, so if your user installed the app, `lp` is already there. macOS and many Linux systems also have a printer command named `lp` (`/usr/bin/lp`): if it comes first on your PATH, call `~/.local/bin/lp` (or put `~/.local/bin` first). In Windows PowerShell 5.1, `lp` is a built-in alias for Out-Printer, so type `lp.exe` there (for example `lp.exe login`); Git Bash, cmd and PowerShell 7 take plain `lp`. If `lp` is not found right after installing, open a new shell or call it by its full path: `~/.local/bin/lp` or `%USERPROFILE%\.logicpacks\helper\bin\lp.exe`. Installing only puts the `lp` command in place. It changes no agent's settings: that happens in step 3, and only after your user has seen the diff. ## 2. Connect to the LogicPacks app Check first: `lp whoami`. If it says LogicPacks is not connected: ```sh lp login ``` This connects the helper to the LogicPacks app running on this computer, with no browser step and no account. If it says the app is not running, ask your user to open the LogicPacks app, then run `lp login` again. If the app is not installed, install it with one line (it also installs `lp`): `curl -fsSL https://logicpacks.io/install.sh | bash` on macOS, or `irm https://logicpacks.io/install.ps1 | iex` in Windows PowerShell; or download it from https://logicpacks.io/download. LogicPacks cloud is coming soon; until then everything runs through the app on this computer. Teams that run their own host add `--host https://their-host`. That path asks your user to approve a device code in a browser: run `lp login --host --start`, show your user the link and code, then `lp login --host --finish` once they approve. ## 3. Add the LogicPacks block to your global instructions First show your user exactly what will change, then write it only after they say yes: ```sh lp setup --agent # prints the unified diff, writes nothing lp setup --agent --yes # after your user approves: writes it ``` Show your user the whole diff from the first command and ask for approval. Do not run `--yes` until they agree. `--dry-run` also shows the diff and never writes. What `lp setup` guarantees: - It only inserts or replaces the text between `` and ``. Every other line of the file stays byte for byte the same, and running it twice changes nothing. - Before it changes a file it copies it to `.lp-backup-` and prints that path. - If the file has a lone or doubled LogicPacks marker, it refuses and changes nothing. - It writes the global file it resolves for your harness (the table below). Leave out `--file` unless your harness is not in the table, and never point it at an instructions file in a folder above the project (such as a parent folder's `.claude/CLAUDE.md`): `lp setup` refuses those. | Harness | id | Global instructions | Project file | Runs `lp` itself | MCP | |---|---|---|---|---|---| | Claude Code | `claude` | `~/.claude/CLAUDE.md` | `CLAUDE.md` | yes | yes | | Codex CLI | `codex` | `~/.codex/AGENTS.md` | `AGENTS.md` | yes | yes | | OpenCode | `opencode` | `~/.config/opencode/AGENTS.md` | `AGENTS.md` | yes | yes | | Gemini CLI | `gemini` | `~/.gemini/GEMINI.md` | `GEMINI.md` | yes | yes | | Command Code | `command-code` | `~/.commandcode/AGENTS.md` | `AGENTS.md` | yes | yes | | Z.ai GLM Coding Plan | `zai` | `~/.claude/CLAUDE.md` | `CLAUDE.md` | yes | yes | | Cursor | `cursor` | settings UI only | `.cursor/rules/logicpacks.mdc` | yes | yes | | VS Code + GitHub Copilot | `copilot` | `~/.copilot/copilot-instructions.md` | `.github/copilot-instructions.md` | yes | yes | | Windsurf | `windsurf` | `~/.devin/global_rules.md` | `AGENTS.md` | yes | yes | | Cline | `cline` | `~/Documents/Cline/Rules/logicpacks.md` | `.clinerules/logicpacks.md` | yes | yes | | Roo Code | `roo` | `~/.roo/rules/logicpacks.md` | `.roo/rules/logicpacks.md` | yes | yes | | Kilo Code | `kilo` | settings UI only | `.kilo/rules/logicpacks.md` | yes | yes | | Zed | `zed` | `~/.config/zed/AGENTS.md` | `.rules` | yes | yes | | Continue | `continue` | settings UI only | `.continue/rules/logicpacks.md` | yes | yes | | Aider | `aider` | `~/.aider.conf.yml (read: ~/.logicpacks/rules/logicpacks.md)` | `CONVENTIONS.md` | asks you | no | | Amp | `amp` | `~/.config/amp/AGENTS.md` | `AGENTS.md` | yes | yes | | Warp | `warp` | settings UI only | `AGENTS.md` | yes | yes | | Kiro | `kiro` | `~/.kiro/steering/logicpacks.md` | `.kiro/steering/logicpacks.md` | yes | yes | | Goose | `goose` | `~/.config/goose/.goosehints` | `.goosehints` | yes | yes | | Qwen Code | `qwen` | `~/.qwen/QWEN.md` | `QWEN.md` | yes | yes | | Crush | `crush` | `~/.config/crush/CRUSH.md` | `AGENTS.md` | yes | yes | | Factory Droid | `droid` | `~/.factory/AGENTS.md` | `AGENTS.md` | yes | yes | | JetBrains Junie | `junie` | `~/.junie/AGENTS.md` | `.junie/AGENTS.md` | yes | yes | | Hermes Agent | `hermes` | settings UI only | `AGENTS.md` | yes | yes | | T3 Code | `t3code` | settings UI only | `AGENTS.md` | yes | yes | | Any other agent | `generic` | settings UI only | `AGENTS.md` | yes | yes | Notes: - **Codex CLI**: An AGENTS.override.md next to it wins over AGENTS.md; LogicPacks writes AGENTS.md. - **Command Code**: commandcode.ai, CLI `cmd` (npm i -g command-code). - **Z.ai GLM Coding Plan**: A model plan, not a harness: it runs inside Claude Code (or OpenCode, Cline, Codex...). Set up the harness you run GLM in; this target is Claude Code's. - **Cursor**: Open Cursor Settings, then Rules (Customize > Rules). Under User Rules, add a rule and paste the LogicPacks block (lp setup-block prints it). User Rules live in the app only. Cursor also reads a root AGENTS.md. - **VS Code + GitHub Copilot**: Personal *.instructions.md files go in ~/.copilot/instructions; AGENTS.md is read too (chat.useAgentsMdFile). - **Windsurf**: Older installs read ~/.codeium/windsurf/memories/global_rules.md; pass --file for that one. - **Kilo Code**: Open ~/.config/kilo/kilo.jsonc (Kilo Code settings). Add the rules file to "instructions", e.g. "instructions": ["~/.logicpacks/rules/logicpacks.md"] (lp setup --agent kilo writes that file). Global rules are listed in kilo.jsonc `instructions`; LogicPacks does not edit JSONC. - **Continue**: Continue's docs name no global rules file; LogicPacks writes .continue/rules/logicpacks.md in this project. For every project, add the block as a rule in your Continue config (Hub or config.yaml `rules:`). - **Aider**: Aider is not an agent loop: it asks you before running a command. Run the lp commands it suggests yourself. - **Warp**: In Warp, open Warp Drive > Personal > Rules (or Settings > AI > Knowledge > Manage Rules). Add a Global rule and paste the LogicPacks block (lp setup-block prints it). - **Goose**: Hints need the Developer extension, which is also what runs shell commands. - **Factory Droid**: The docs say Droid looks in ~/.factory/; the exact global file name is unverified. - **JetBrains Junie**: Older projects use .junie/guidelines.md, which Junie still reads. - **Hermes Agent**: Hermes has no global instructions file (~/.hermes/SOUL.md is persona only). LogicPacks writes the block into this project's AGENTS.md, which Hermes reads. - **T3 Code**: T3 Code runs the agents you have installed (Codex, Claude Code, OpenCode...) and has no file of its own. Set up each of those (lp setup --agent codex, --agent claude...); LogicPacks also writes this project's AGENTS.md. - **Any other agent**: Pass the file your agent reads its instructions from: lp setup --agent generic --file . Without --file, LogicPacks writes this project's AGENTS.md, which most agents read. - Any other harness: `lp setup --agent generic --file `. If you would rather edit the file yourself, `lp setup-block` prints the block; paste it with its markers. - To also write the project file (for example `AGENTS.md`), add `--project`. The block is: ```markdown ## LogicPacks Team work runs through the `lp` command (add `--json` to parse its output). If `lp` is the printer command (macOS, Linux), call `~/.local/bin/lp`; in Windows PowerShell 5.1 type `lp.exe` (`lp` is an Out-Printer alias there). If it is missing, follow https://logicpacks.io/setup.md. - At the start of a session and after long tasks, run `lp inbox` and tell the user what is new (messages, questions from other agents, conflicts, drift, review requests). - "share this project" / "create a shared session" -> `lp share`, then give the user the invite link. - "add my friend X" / "invite X" -> `lp invite ` (this project only; `--workspace` for every project); "remove X" -> `lp remove `; "turn off the invite link" -> `lp revoke-invite`. - "what is X working on" -> `lp who X`; "who is working right now" -> `lp agents` (`--workspace` for every project). - "tell X ..." -> `lp send X ""`. - To coordinate with a teammate's agent, use `lp ask ""` and `lp reply ""`; check `lp inbox` for questions from other agents and answer them. - "request assistance" / "get an expert" -> `lp assist request ""`. - "set goals" / "ask X to do Y by Friday" -> `lp goals plan "" ": by " ...`; your user confirms with `lp goals confirm `. Milestones for you arrive in `lp inbox`: `lp goals start|done `. - "save this as a pack" -> `lp packs save ""`; `lp packs list`, `lp packs use `. - Before you edit shared files, run `lp who `; if a teammate is on it, tell the user first. - On a conflict or negotiation in `lp inbox`, answer with `lp reply `; when it needs a person, ask the user, then `lp decide mine|theirs|both`. - If you cannot run shell commands, ask the user to run the `lp` command, or use the `logicpacks` MCP tools if they are set up. - Everything else: `lp help agents` or https://logicpacks.io/setup.md. `lp uninstall` removes this block. ``` ## 4. Share the project If your user asked you to share the project they are in: ```sh lp share ``` It pushes the folder to its git remote (it creates a private GitHub repository with `gh` when there is none), registers it on the LogicPacks host and prints an invite link. The link grants access to this project only; tell your user that, and give them the link. To add a teammate by email: `lp invite friend@example.com` (also this project only). `lp invite --workspace` gives a link to every project in the workspace: use it only when your user asks for that. ## 5. Every session from now on - Run `lp inbox` at the start of a session and after long tasks. Tell your user about anything new. Each item lists the commands to act on it. - Map your user's requests to commands with the block above. The full list: ```text LogicPacks for agents. Every command prints readable text; add --json for machine output. lp login [--cloud | --host ] [--start | --finish] connect to the LogicPacks app on this computer (no browser step); --cloud or --host sign in to a hosted one with a device code lp whoami who you are signed in as, and on which host lp projects the projects you can open lp share [--title ] [--workspace] share this folder: push it to its git remote (GitHub) and register it on the host; prints an invite link lp invite [] [--workspace] add someone to this project, or print the invite link to send them lp join accept an invite link from a teammate lp members who is in this project lp who [person|file|topic] who is working on what in this project lp agents [--workspace] active agents and sessions: person, harness, thread, current task lp send "" message a teammate's thread; no reply expected lp ask "" [--project ] ask a teammate's agent; the reply comes back to your lp inbox lp reply "" answer a message, question or negotiation from lp inbox lp decide accept|mine|theirs|both record your user's decision on a merge negotiation lp inbox [--follow] [--all] messages, questions, conflicts, negotiations, drift and review requests for you lp editing tell teammates which files you are changing (lp who shows it; overlaps raise a notice) lp done dismiss an inbox item you handled lp assist request "" [--expert ] Request assistance: bring in an expert (beta) lp assist experts | list | end the expert catalog, this project's requests, end a request lp packs save "" | list | use [--to ] save this session as a pack, list packs, hand one to a thread lp goals plan "" ": [by ]"... draft a goal with milestones for people or experts (beta); nothing is sent yet lp goals list | show | confirm | cancel see goals; confirm a draft (sends the milestones) once your user agrees lp goals start|blocked|done [--note ""] report on a milestone assigned to you lp goals review | decide [--note] drift and review requests waiting for your user, and their decision lp remove [--project-only] [--dry-run] take someone's access away (this project and its workspace) lp revoke-invite [] [--off] [--workspace] turn off this project's invite link (and print a new one), or cancel one person's invite lp stream --status | --mcp-details on|off what this folder's sessions send to the project; MCP details are off by default lp setup --agent [--file ] [--project] [--yes] show the diff for the LogicPacks block; --yes writes it (with a backup) lp setup-prompt [--installed] print the prompt to paste into any agent lp install [--dry-run | --yes] optional: Claude Code hooks, background watcher, MCP for Claude Code and Codex lp pause | lp resume | lp unlink | lp link stop streaming sessions (everything, or just this folder) and turn it back on lp status | lp watch connection and queue; the background session streamer lp mcp the optional MCP server (stdio) lp uninstall [--dry-run] [--agent ] remove every LogicPacks block, rules file, hook and MCP registration Setup guide: https://logicpacks.io/setup.md ``` ## Optional: MCP Only if your user asks for it. Show the diff first with `lp install --dry-run`, then `lp install --yes` after they approve (each changed file is backed up first). If your user wants MCP tools as well, `lp install` registers the `logicpacks` MCP server (`lp mcp`, stdio) with Claude Code and Codex (and adds the Claude Code session hooks and the background watcher that stream sessions into the project). Other harnesses can add it by hand: command `lp`, args `["mcp"]`. The tools are a small subset of the CLI (who is working on what, send, ask, reply, conflicts, notes, packs). The CLI is the main way. ## What streams to the project Only after `lp install --yes`, Claude Code and Codex sessions in a folder that belongs to a project stream into it, so teammates see what the agents are doing: - prompts and replies, with secrets redacted (API keys and tokens such as sk-, ghp_, xox, AKIA, JWTs, private key blocks, password= and token= pairs, secret-looking KEY=VALUE lines, credentials in connection strings); - built-in tool calls (commands, edits, reads, searches): up to 300 characters of input and 1,500 of result, redacted the same way; - MCP tool calls as names only, such as "mcp: whatsapp.send", with no input or result. `lp stream --mcp-details on` in a folder includes them (still redacted); - never the content of .env*, *.pem, *.key, id_* or ~/.ssh files. `lp stream --status` shows what the current folder sends. `lp unlink` keeps one folder private, `lp pause` stops all streaming on this computer, `lp resume` and `lp link` turn it back on. ## What gets installed where The helper (step 1), macOS and Linux: - ~/.logicpacks/helper/bin/lp (the CLI: one standalone executable, no Node needed) - ~/.local/bin/lp (a symlink to it) - ~/.logicpacks/helper/config.json (host, token, device name; mode 600), setup.json (what lp setup wrote), queue/ (sessions waiting to upload) The helper (step 1), Windows: - %USERPROFILE%\.logicpacks\helper\bin\lp.exe (the CLI: one standalone executable, no Node needed), plus bin\lp.cmd (cmd) and bin\lp (Git Bash) that start it - the user PATH gains %USERPROFILE%\.logicpacks\helper\bin - %USERPROFILE%\.logicpacks\helper\config.json, setup.json, queue\ Step 3: lp setup --agent : one marked block in that harness's global instructions file (see the table), or one rules file of its own (Cline, Roo, Kiro, Cursor .mdc); with --project or for UI-only harnesses, the project's file (for example AGENTS.md). Only with `lp install --yes`, macOS and Linux: - ~/.claude/settings.json: hooks SessionStart, UserPromptSubmit, PostToolUse, Stop that run `lp hook ` (marked # logicpacks-helper) - ~/.claude.json: mcpServers.logicpacks (stdio: lp mcp --source claude) - ~/.codex/config.toml: [mcp_servers.logicpacks] table - macOS: ~/Library/LaunchAgents/io.logicpacks.helper.plist (runs lp watch); Linux: ~/.config/systemd/user/logicpacks-helper.service Only with `lp install --yes`, Windows: - %USERPROFILE%\.claude\settings.json hooks and %USERPROFILE%\.claude.json MCP entry, as on macOS - %USERPROFILE%\.codex\config.toml [mcp_servers.logicpacks] - %APPDATA%\Microsoft\Windows\Start Menu\Programs\Startup\logicpacks-helper.vbs (runs lp watch hidden) Backups: .lp-backup- next to each file before it changes. ## Removing access and uninstalling - `lp remove ` takes someone's access away (this project and its workspace; `--project-only` for just the project). `--dry-run` says what it would do. - `lp revoke-invite` turns off the current invite link and prints a new one. ```sh lp uninstall --dry-run # the exact diff, writes nothing lp uninstall ``` It removes every LogicPacks block and rules file `lp setup` wrote, the Claude Code hooks, the background watcher and the MCP registrations, and leaves the rest of each file exactly as it was. `lp logout` forgets the token. ## Before you run it What this setup installs and changes, how to back up first, what teammates can see and how to remove access: https://logicpacks.io/safety.md ## Code with a friend If you only want to work with one other person, the app (Mac or Windows) is enough: 1. Each of you installs the app (Mac or Windows) from https://logicpacks.io/download and has Claude Code or Codex installed and logged in. If the CLI is missing, the app shows the install command and tells you to log in in your terminal. 2. The host adds a project in the app: an empty folder, or an existing folder or Git repository. 3. The host adds the other person and shares the project. Teammates connect and sync to the host's laptop while it is on. 4. Each of you works in your own thread with your own agent on your own login, and you both see who is working on what.