Technology RadarTechnology Radar

Archon

workflowagentopen-sourceframework
Trial

Archon is an open-source AI coding workflow engine that encodes development processes — planning, implementation, testing, code review, PR creation — as YAML-defined DAGs. Each workflow mixes deterministic nodes (bash scripts, test runners) with AI nodes (LLM prompts), runs in an isolated git worktree, and can be triggered from CLI, Web UI, Slack, Telegram, Discord, or a GitHub webhook.

The Core Idea

Archon's positioning mirrors how GitHub Actions standardized CI/CD: "what Dockerfiles did for infrastructure and GitHub Actions did for CI/CD — Archon does for AI coding workflows." Instead of prompting an AI agent freehand, you encode your best process into a reusable YAML workflow. The AI fills in the intelligence at each node; the structure — phases, validation gates, loop conditions, approval steps — remains deterministic and owned by the team.

This makes it a direct implementation layer for the Harness Engineering pattern: the workflow file is the harness.

Architecture

Platform Adapters (CLI · Web UI · Slack · Telegram · Discord · GitHub)
         │
    Orchestrator (message routing, context management)
         │
    ┌────┴────────────────────────────────┐
    │                                     │
Command Handlers          Workflow Executor (DAG runner)
                                │
                     AI Assistant Clients
                     (Claude Code, etc.)
                                │
              SQLite / PostgreSQL (7-table schema)
              codebases · conversations · runs · worktrees

Each workflow run gets its own git worktree, preventing environment contamination between concurrent runs. The database tracks full execution history, enabling reproducible re-runs and post-hoc debugging.

YAML Workflow Structure

A workflow defines nodes with typed roles and dependency edges:

nodes:
  plan:
    type: ai
    prompt: "Analyze the issue and produce a step-by-step plan."
  implement:
    type: loop
    depends: [plan]
    prompt: "Implement the next step of the plan."
    condition: "until: tests_pass"
    fresh_context: true
  test:
    type: bash
    depends: [implement]
    command: "npm test"
  review:
    type: ai
    depends: [test]
    prompt: "Review the implementation for correctness and style."
  approve:
    type: approval         # interactive human gate
    depends: [review]
  create_pr:
    type: bash
    depends: [approve]
    command: "gh pr create --fill"

Key node types: ai (LLM prompt), bash (deterministic command), loop (iterative AI work with a completion condition), approval (human gate), script (run TypeScript or Python scripts inline — added in v0.3.3). Archon ships 17 built-in workflows covering common tasks.

Why Trial

  • ~19.9k GitHub stars (up from 15.5k in mid-April 2026) — continued organic growth above the Assess threshold
  • MIT licensed with no commercial lock-in
  • TypeScript (97.6% of codebase) — auditable and contributor-friendly
  • Hybrid deterministic + AI execution is the right architectural answer to the chaos problem of freehand prompting
  • Worktree isolation enables safe parallel runs — the same insight behind Composio AO and Stripe's Minions
  • Multi-platform triggers (Slack, Discord, GitHub) bring it into existing team workflows
  • Env-leak gate (v0.3.0): scans .env files for sensitive keys (API keys, secrets) and blocks workflows from silently re-injecting them into AI subprocess contexts — a production-safety feature missing from most orchestrators
  • Native Codex support (v0.3.3): auto-resolves Codex binary alongside Claude Code

The TypeScript rewrite is now shipping stable patch releases (v0.3.9 as of April 22, 2026). The initial rewrite caveat no longer applies.

When Archon Makes Sense

  • You want to standardize how AI coding agents execute across your team
  • You're already practicing harness engineering and want a concrete tool layer
  • You need repeatable, auditable AI coding runs (for compliance or onboarding)
  • You want to trigger AI coding workflows from Slack/GitHub without a custom integration layer

Not Right For

  • One-off AI coding tasks where YAML workflow overhead isn't justified
  • Teams that need enterprise auth, SSO, or audit logging today (not yet in scope)
  • Building custom agent logic (use LangGraph, CrewAI, or OpenAI Agents SDK instead)

Key Characteristics

Property Value
Language TypeScript
License MIT
GitHub coleam00/Archon
Stars ~19.9k (April 2026)
Latest release v0.3.9 (April 22, 2026)
Website archon.diy
Built-in workflows 17
Storage SQLite (default) / PostgreSQL
First version Python-based (archived); TypeScript rewrite current

See Also

Sources