ORG-2 is a local-first desktop harness that runs coding agents and turns every session into a replayable trajectory, so a team can trace any shipped line of code back to the decision that produced it.

What it is

ORG-2 is a desktop application built with Rust and Tauri (with a TypeScript front end) that acts as a system of record for how agents build software. It ships a native Rust harness alongside launchers for more than 20 coding-agent CLIs — including Cursor CLI, Claude Code, Codex, Kiro CLI, GitHub Copilot, OpenCode, DeepSeek Harness, and Antigrav — and it keeps execution under 100MB on disk by running locally rather than in a hosted workspace.

The problem it addresses is fragmentation: the README points out that Jira sees only tickets, Codex sees only its own sessions, GitHub sees only committed lines, and Amplitude sees only metrics, which meant answering why a piece of code exists required stitching systems together by hand. That becomes untenable at agent speed, where code written on Monday is legacy by Friday. ORG-2 builds the record automatically, linking what the human asked for, what the agent understood, and what the agent actually did, and it backfills sessions run in other tools from their history so work that never touched the app still enters the record.

Key capabilities

• Replays tool calls, messages, file edits, and command output as a synchronized review timeline — described in the README as "Tool calls, now as videos" — covering the native Rust harness and 15+ CLI agents.
• Provides "AI blame, not just Git blame", tracing a line back to the agent session, tool call, and decision that drove the change rather than stopping at who committed it.
• Loads and manages agent sessions from 10+ apps and CLIs in one place, scanning history and inspecting subagents without switching applications.
• Offers Design Mode, which inspects live pages in a native WebKit browser and sends an element's exact page context directly to the agent for a fix.
• Provides a full dev workspace containing a terminal, source control management, Git history tracing, and pull request review, plus GUI, CLI, LSP, timeline, and database tooling.
• Runs resource-aware execution that reacts to CPU, RAM, and human attention availability, with scheduling and auto-started sessions so agents can run overnight.
• Maintains cross-session memory, cross-agent knowledge sharing, and shared workspace state, with a daily activity timeline showing duration, code changes, and priorities.

Who uses it and how

• Teams share sessions across devices and teammates, reviewing the full agent trajectory rather than only the resulting diff, and leaving comments in context.
• Developers who juggle several agent CLIs load history from all of them into one workspace instead of switching apps for each source.
• Solo self-hosters run coding agents locally with their existing API keys and agent subscriptions, keeping execution on their own machine.
• Groups coordinating humans and agents use the work-in-progress issues and projects management surfaces for org-level alignment and accountability.
• Testers use agent-powered GUI end-to-end testing for supervised self-evolution.

Getting started

Installers are published as release assets for macOS Apple Silicon, a Windows installer and MSI, and Linux AppImage and DEB packages, available from the project's homepage at org2.dev.

How it compares

Unlike Jira, GitHub, or Amplitude — each of which sees only tickets, only committed lines, or only metrics — ORG-2 is positioned as the single place where sessions, decisions, and diffs are joined into one traceable record. Within the agent tooling it overlaps with, such as Cursor CLI, Claude Code, and Codex, it acts as the harness and review layer that runs those CLIs rather than replacing them.

When to use it — and when not to

Session collaboration and group issue workflows are delivered through self-hosted Supabase, so an operator must stand up and maintain that service, and both that feature and org-level alignment management are still marked WIP alongside 77 open issues. The AGPL-3.0 licence suits self-hosted and open-source use but warrants a look before adopting it in a closed-source product, and the project is explicitly described as an experiment in human/agent organizations rather than a finished offering.

pnpm install
pnpm run download:sidecars
pnpm run tauri:dev