What is Archon?
Archon has pivoted. It began in 2025 as a Python task-management and RAG server for coding assistants; that version is preserved on the archive/v1-task-management-rag branch. The current Archon, rewritten in TypeScript on Bun, is a workflow engine: you describe a process as a YAML graph of agent prompts, bash commands, loops and human approvals, and it runs Claude Code, Codex or Pi through their own CLIs and SDKs, each repository run in its own git worktree.
The bundled sdlc pack is a working pipeline from issue to merged pull request: an intake workflow triages each issue, a build queue plans, implements, reviews and waits for CI, and a merge queue lands pull requests that pass its gate. Questions a run cannot settle arrive as single attention items in a web console, the terminal, Slack or Telegram. State lives in SQLite or PostgreSQL. It is MIT-licensed, pre-1.0 and releases fast: v0.14.0 shipped on 7 November 2026.
Is Archon worth it?
Archon is the most complete open-source answer to a problem every team using coding agents hits: the same request gets a plan one day and none the next, tests sometimes, the pull-request template rarely. Instead of a longer prompt, Archon makes the process code. A node can loop on a prompt until_bash a test command passes, with a max_iterations cap and fresh context each round; another can pause the run for an approval form; conditions route on the answer. The agent supplies judgement inside each step, and the YAML decides the order.
The bundled sdlc pack shows how far that goes. It triages issues, claims the next ready one, plans, implements, reviews through specialist "lenses", fixes findings and waits for CI, then a merge queue lands pull requests whose review verdict and head-commit CI are green. The maintainers report that Archon runs built 363 of the 524 pull requests merged into its own repository between 1 and 19 October 2026, a self-reported figure but a strong signal that the pack carries real load.
It is a harness, not an agent. Archon drives Claude Code, Codex and Pi through your existing accounts and records an event log, transcript and cost for every run in your own database, which makes it a better fit than a hosted agent platform for a team that wants an audit trail. Compared with spec-driven toolkits such as Spec Kit, which shape what the agent writes, Archon controls when it is allowed to stop.
The catch: this is pre-1.0 software moving at speed. There have been releases from v0.11 to v0.14 since late September, and v0.14.0 removed the sdlc pack's gh path outright, so it now needs a separate forge plugin to reach GitHub. Anonymous telemetry to PostHog is on by default, covering workflow shapes, models, token totals and cost but not code or prompts, until you set ARCHON_TELEMETRY_DISABLED=1. And parallel worktree runs with test loops spend agent tokens quickly; the per-run cost log is there because you will want it.
Our take: Archon earns its place when you already know your team's delivery process and want an agent held to it. For one developer who just wants agents in parallel, Claude Squad is a far smaller commitment.
Where it earns its place
A six-person product team on GitHub using Claude Code
Turn small, well-written issues into reviewed pull requests without a developer babysitting each one
Install the sdlc pack and forge-github plugin, then let the build queue claim ready issues; runs pause only for scope questions or approvals. The software is free; the bill is Claude Code usage, which the per-run cost log makes visible.
A tech lead tired of agents skipping the tests
Make every agent change pass the test suite before anyone looks at it
Write a workflow whose implement node loops with until_bash on the test command and max_iterations: 10, followed by an approval wait. The agent cannot hand back work while the tests fail.
A compliance-minded engineering manager at a fintech
Show auditors who approved each AI-made change and what it cost
Self-host Archon on PostgreSQL; every run stores its event log, transcript, approval decisions and cost. Set ARCHON_TELEMETRY_DISABLED=1 before rollout.
Before you commit
- Setup
- Install and sign in to Claude Code or Codex, then curl -fsSL https://archon.diy/install | bash (PowerShell and Homebrew installers also exist); archon workflow run archon-investigate in any git repository gives a first result in minutes. Docker and server installs are in the docs.
- Integrations
- Agent providers Claude Code, Codex and Pi; GitHub through the forge-github plugin; Slack, Telegram, Discord and GitHub webhooks for starting and answering runs; a web console via archon serve; plugins install from any GitHub repository with archon plugin add.
- Data and privacy
- Runs locally or on your server with SQLite or PostgreSQL; code and prompts go only to your agent provider. Anonymous telemetry to PostHog is on by default (workflow shapes, provider and model ids, token totals, cost; no code, prompts or file paths) and off in CI; disable it with ARCHON_TELEMETRY_DISABLED=1 or DO_NOT_TRACK=1. MIT licence.
- Cost at scale
- Archon itself is free with no paid edition; cost is agent usage, multiplied by loop iterations and parallel worktree runs. A self-hosted server needs only a small VM, since the agents do the heavy work through their providers.
- Limits
- Pre-1.0: v0.14.0 broke the sdlc pack's GitHub path and requires the forge plugin; 153 open issues and pull requests; GitHub is the only maintained forge, and runs inside a container cannot reach host commands such as the forge check.
What’s good
- Test loops and approval gates enforced by the engine, not the prompt
- Bundled pipeline from issue to merged pull request
- Runs Claude Code, Codex and Pi on your existing subscriptions
- Event log, transcript and cost recorded per run in your database
- Each run isolated in its own git worktree
What’s not
- Pre-1.0 with breaking changes between minor releases
- Telemetry on by default
- Loops and parallel runs multiply agent token spend
- GitHub is the only maintained forge; Slack the only maintained chat plugin
- Original task-manager and RAG version abandoned to an archive branch
Best for
Engineering teams that have a delivery process they trust and want coding agents to follow it every time, with a human approving only the decisions that need one.
Archon alternatives
Archon: common questions
- Is Archon worth it?
- Archon turns "please remember to run the tests" into a loop the agent cannot leave until they pass, which is the right idea, shipped by a project still breaking its own interfaces every few weeks. We rate it 4.0 out of 5.
- How much does Archon cost?
- Free under MIT. No paid edition; you pay for the Claude Code or Codex subscriptions or API usage the workflows consume.
- What are the downsides of Archon?
- Pre-1.0 with breaking changes between minor releases; telemetry on by default; loops and parallel runs multiply agent token spend; gitHub is the only maintained forge; Slack the only maintained chat plugin; original task-manager and RAG version abandoned to an archive branch.
- Who is Archon best for?
- Engineering teams that have a delivery process they trust and want coding agents to follow it every time, with a human approving only the decisions that need one.
- How much does Archon cost for a team or in production?
- Archon itself is free with no paid edition; cost is agent usage, multiplied by loop iterations and parallel worktree runs. A self-hosted server needs only a small VM, since the agents do the heavy work through their providers.
- Is Archon safe for company data?
- Runs locally or on your server with SQLite or PostgreSQL; code and prompts go only to your agent provider. Anonymous telemetry to PostHog is on by default (workflow shapes, provider and model ids, token totals, cost; no code, prompts or file paths) and off in CI; disable it with ARCHON_TELEMETRY_DISABLED=1 or DO_NOT_TRACK=1. MIT licence.
- What are the best Archon alternatives?
- The closest alternatives we have reviewed are Spec Kit, Superpowers, Claude Squad and OpenHands.
Answers checked Nov 9, 2026.
Make Archon? Show this rating on your site
