Introduction

ORG-2 is an open-source desktop app for working with coding agents. It runs locally, stores your data on your own machine, and is built on Rust and Tauri rather than Electron. What sets it apart from a chat sidebar or a terminal agent is that every session is recorded as a structured, replayable trajectory — so you can watch an agent work, audit what it did afterwards, and hand that record to a teammate.

The project describes itself as an Agentic Development Environment (ADE) rather than an IDE. An IDE assumes the human is the primary actor and optimizes for human typing speed. An ADE assumes humans and agents are both first-class actors, and optimizes for delegation, observation, and review.

The problem ORG-2 exists to solve

Agent capability is improving faster than the tooling around it. In most agent IDEs:

• Work is opaque. Tool calls collapse, reasoning is hidden, and you see a final answer without the process that produced it. An agent you cannot audit is an agent you cannot trust.
• Work is unreviewable. A session that produced correct output but left no legible trace cannot be debugged, improved, or handed off. Onboarding someone to a session they did not run is close to impossible.
• Work is uncoordinated. Context lives inside one person's chat history. Two agents duplicate effort, memory is siloed per tool, and nothing at the team level tracks who — human or agent — is accountable for what.
• Setups do not travel. Switching tools means rebuilding your MCP configuration, your skills, and every accumulated piece of project context from scratch.

ORG-2 treats these as the actual product problem, not a side effect.

What makes it different

Alongside this the app ships GUI, CLI, terminal, Git, browser, LSP, timeline, and database tooling; scheduling and auto-started sessions so agents can keep working overnight; and resource-aware execution that can react to CPU, RAM, and whether you are actually at the keyboard.

Note: Org-level alignment surfaces and session collaboration are marked work-in-progress in the project README. Expect them to move faster and change more than the session and replay layers.

Bring your own agents and keys

ORG-2 does not resell inference and does not lock you into one model vendor. You connect what you already pay for:

• Direct API providers — Anthropic, OpenAI, Google Gemini, DeepSeek, Groq, xAI, Zhipu, DashScope, Moonshot, MiniMax, OpenRouter, ZenMux, Azure OpenAI, Azure-hosted Anthropic, and self-hosted vLLM.
• CLI-based coding agents — Claude Code, Codex, Cursor CLI, Gemini CLI, GitHub Copilot, Kiro, Kimi CLI, and OpenCode.
• External agent sessions — Cursor IDE activity is imported and rendered as a trajectory alongside native sessions.

Native ORG-2 agents are written in Rust and run in the same local process as the rest of the app. See API keys and models and Agents.

Who it is for

• Developers who review agent output seriously — people who want a merge-time audit trail, not just a diff that appeared from nowhere.
• Teams delegating work to agents — where the hard part is coordination, shared context, and knowing who owns what.
• People shipping small products — the design goal is to help you get an app out the door and into other people's hands, not only to help you write code faster.
• Anyone who wants their setup to be portable — your keys, skills, MCP servers, and memories stay yours and stay local.

Open core

ORG-2 is open core, and the split is deliberate:

• The desktop app is open source under AGPL-3.0-or-later. The full source, including the Rust agent runtime, lives in the public repository.
• Self-hosted Supabase is free. Team sync — orgs, members, shared session metadata, group chat — runs against a Supabase project you own. The app can generate the setup SQL for you.
• ORG-2 Cloud is the optional managed backend. It hosts the same sync surface plus account, billing, and sharing features for people who do not want to run their own Postgres. See ORG-2 Cloud.

Nothing about the local, single-player experience requires an account. Contributions are welcome and require a signed CLA; see CONTRIBUTING.md in the repository.

Platform support

Two capabilities depend on optional native sidecars downloaded at build time: Browser Use uses agent-browser, and Computer Use uses peekaboo. Computer Use is macOS only. If a sidecar is missing, the related capability falls back to PATH or stays unavailable until you run pnpm run download:sidecars.

How to read these docs

Work through them in roughly this order.

Get running

• Install — download, first launch, and building from source.
• Quickstart — add a repo, add a key, run your first session.
• Concepts — the vocabulary the rest of the docs assume.

Daily use

• API keys and models — connect providers and CLI agents.
• Agents — built-in agents, CLI agents, custom agents, and agent teams.
• Sessions — starting, resuming, and steering a session.
• Replay — reading a trajectory after the fact.
• Tools — what agents can actually do, and the policies that gate it.

Extend the agent

• MCP servers — connect external tool servers.
• Skills — add SKILL.md capabilities.
• Memory — learnings, workspace memory, and consolidation.
• Scheduling — recurring and auto-started sessions.

Work with other people

• Projects — projects and work items.
• Orgs — grouping projects, and agent teams.
• Collaboration — sharing sessions with teammates.
• ORG-2 Cloud — the managed backend.

Operate it

• Security — key storage, sharing tiers, and redaction.
• Troubleshooting — when something does not work.

Next steps

• Install ORG-2 and get the desktop app running.
• Read the core concepts before you configure anything.
• Follow the quickstart to run and replay your first session.