org2AI/ORG2: a session record for coding agents, not another autocomplete
System of record for how agents build software: Built-in rust harness & 20+ CLIs.
At a glance
- What is it?
- ORG2 is a Tauri desktop app with a built-in Rust agent harness and launchers for 20+ coding-agent CLIs. It records every session as a replayable trajectory, ingests history from other tools, and links shipped lines back to the decisions that produced them.
- Who is it for?
- ORG2 fits teams that already run several coding-agent CLIs and want one replayable record of what each agent did, plus a native harness for the sessions they start inside the app. Skip it if you need a stable, documented review workflow today: the README marks the org-level alignment surfaces and the self-hosted Supabase collaboration path as WIP.
- Can I use it commercially?
- Yes, with strict conditions. AGPL-3.0 is a network copyleft licence: if people use a modified version over a network, for example as a hosted service, you must offer them its source code under the same licence.
- Is it still maintained?
- Yes. The repository last received commits 4 days ago.
- What is it written in?
- Mainly TypeScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 14, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
The gap ORG2 is built around: agents leave no reviewable record
The README frames the problem in terms of what each existing system can see. Jira sees tickets. Codex sees its own sessions. GitHub sees committed lines. Amplitude sees metrics. None of them sees the sequence of tool calls, file edits and command output that produced a commit. The project's own summary of why that mattered less before: "That was survivable when humans wrote the code. At agent speed it isn't."
The intended user is a team running coding agents as part of normal work, not a solo developer trying a chatbot once. The pitch is that a session becomes a trajectory a teammate can replay like a video, comment on in context, and trace back to a shipped line. That is a different product category from an editor plugin that suggests code. ORG2 is closer to a flight recorder for agent work, and the README says so directly: it calls itself "not just another AI coding tool" but an experiment in human and agent organizations.
Two runtimes in one app: a native Rust harness and launchers for external CLIs
The architecture visible in the README has two halves. The first is a built-in Rust harness, described as fast, token-saving and customizable, running with your existing API keys and agent subscriptions. The second is a launcher layer for external agent CLIs. The supported list is split into GUI plus TUI entries (ORG2 itself, Cursor CLI, Claude Code, Codex, Kiro CLI, GitHub Copilot, OpenCode, Antigravity) and a TUI-only group that begins with Kimi Code CLI and is truncated in the README excerpt.
The record layer is what ties them together. Sessions started in the native harness are captured directly. Sessions that ran in other tools are, in the README's wording, "ingested and backfilled from their history, so the record covers work that never touched the app." Replay keeps messages, tool calls, file edits and command output synchronized in one timeline. The repository layout matches this: src-tauri/ for the Rust side, src/ and apps/ for the TypeScript front end, tools/ for diagnostics, and a scripts/tools/download_sidecars.py step that suggests external binaries are fetched at setup time rather than vendored.
Two features go beyond replay. AI blame extends Git blame by tracing a line back to sessions, tool calls and decisions. Design Mode inspects live pages in the native WebKit browser and sends an element's page context to the agent. The second is a concrete, narrow mechanism; the first depends entirely on how complete the ingested history is, which the README does not quantify.
Installing ORG2 and replaying your first session
The README points at prebuilt installers rather than a package-manager install. The download links are published under the latest GitHub release: a macOS Apple Silicon DMG, a Windows setup executable and MSI, and Linux AppImage and DEB assets. There is no documented homebrew, winget or apt path, so the release page is the distribution channel.
If you want to build from source instead, the repository is a pnpm workspace. The package.json defines the front-end dev server and the Tauri launcher as separate scripts, and the postinstall hook runs scripts/setup/postinstall.mjs, which is where sidecar downloads are likely triggered.
pnpm install
pnpm dev:frontend
pnpm tauri:devpnpm dev:frontend starts the webpack dev server; pnpm tauri:dev launches the desktop shell through scripts/dev/tauri-launcher.cjs. A lighter front-end variant exists as dev:frontend:light, and an rspack variant as dev:frontend:rspack, if the default build is slow on your machine.
For a first real use, the workflow the README describes is: run an agent session inside the app or let ORG2 ingest one from a CLI you already use, then open that session as a trajectory. You should see messages, tool calls, file edits and command output on one synchronized timeline, with the option to leave a comment at a point in the timeline rather than on a diff. The README does not document a rollback or undo path for a session, and it does not describe how to delete an ingested session, so treat the record as append-oriented until you confirm otherwise in the app.
Where ORG2 is the wrong tool
The README marks two of its larger claims as work in progress: org-level alignment surfaces for coordinating humans, agents, goals and accountability, and session collaboration with group issue workflows via self-hosted Supabase. Both are labeled WIP. If cross-teammate review is the reason you are evaluating ORG2, you are evaluating an unfinished part of it, and the README does not document what a self-hosted Supabase deployment requires.
The record is also only as good as the ingestion. The README claims backfill from other tools' history, but it does not state which of the 20+ supported CLIs have complete history, which have partial history, or what happens when a session ran on a machine that never had ORG2 installed. A trajectory assembled from partial history can look authoritative while missing the tool call that explains a change, which is worse than no record for review purposes.
Finally, the scope is broad by design: terminal, source control, Git history, pull requests, LSP, browser and database tooling all sit inside the app. That is a lot of surface for a project whose last push was on 2026-08-26. If you want a small tool that does one thing, this is the opposite of that.
How ORG2 differs from a session viewer like Claude Code or Codex alone
The closest alternative is not another agent. It is using each agent's own session history. Claude Code and Codex both keep records of their own sessions, and the README acknowledges this: "Codex sees only its own sessions." That is the actual difference in approach. A single-vendor session log is complete for that vendor and blind to everything else, and it is not built for a second person to review.
ORG2's bet is aggregation plus review. One record across the native harness and the external CLIs, one timeline format, and a replay surface designed for a teammate rather than the person who ran the session. The trade-off is real: you gain coverage across tools, and you take on a desktop application, a Rust harness and an ingestion layer as dependencies. If your team has standardized on exactly one agent CLI and never reviews agent work as a group, the aggregation buys you little.
A second comparison is Git blame itself. Git blame answers who touched a line and when. AI blame answers which session, which tool calls and which decisions produced it. That is a strictly larger question, and it depends on data Git does not have.
Licence and the cost of keeping up
ORG2 is AGPL-3.0. That matters more here than for a library you import, because ORG2 is a desktop application that also offers a self-hosted Supabase path for collaboration. If you modify it and let users interact with it over a network, the AGPL's network clause is the part to have a lawyer read, not this article. The README does not state a separate commercial licence or an exception, so the AGPL is the only terms given.
Upgrade cost is shaped by the dependency on external CLIs. The supported list includes Cursor CLI, Claude Code, Codex, Kiro CLI, GitHub Copilot, OpenCode, Antigravity and Kimi Code CLI, among others. Each of those can change its configuration format or session storage independently of ORG2. The release cadence visible in the repository is fast: v1.2.5 on 2026-08-09, v1.2.6 on 2026-08-17, v1.3.0 on 2026-08-26. Frequent releases plus a moving ingestion target means you should expect to update ORG2 when an upstream CLI changes, and the README does not describe a compatibility matrix or a deprecation policy for CLI support. The last push to the develop branch was on 2026-08-26.
Editorial conclusion
ORG2 fits teams that already run several coding-agent CLIs and want one replayable record of what each agent did, plus a native harness for the sessions they start inside the app. Skip it if you need a stable, documented review workflow today: the README marks the org-level alignment surfaces and the self-hosted Supabase collaboration path as WIP. Before committing, verify the AGPL-3.0 obligations for your distribution model, and confirm that the CLI you depend on is in the supported list rather than only in the truncated part of the README.
Frequently asked questions
What is org2AI/ORG2?
It is a local-first desktop application, built with Rust and Tauri, that runs coding agents and records every session as a replayable trajectory. It ships a native Rust harness plus launchers for 20+ agent CLIs, and it ingests sessions that ran in other tools so the record covers work done outside the app.
How do I install ORG2?
The README links prebuilt installers from the latest GitHub release: a macOS Apple Silicon DMG, Windows setup EXE and MSI, and Linux AppImage and DEB. Building from source uses the pnpm workspace scripts pnpm install, pnpm dev:frontend and pnpm tauri:dev.
Which coding agents does ORG2 support?
The README lists ORG2's own Rust harness, Cursor CLI, Claude Code, Codex, Kiro CLI, GitHub Copilot, OpenCode and Antigravity in the GUI plus TUI group, with a further TUI-only group beginning with Kimi Code CLI. The README excerpt is truncated, so the full list should be checked in the repository.
What licence does ORG2 use?
ORG2 is licensed under AGPL-3.0. The README does not mention a commercial licence or an exception, and the project also offers a self-hosted Supabase path for collaboration, which is the part most likely to interact with the AGPL network clause.
Community notes