Archon: Repeatable AI Agent DAG Workflows
Archon packages AI coding workflows into YAML DAGs for parallel execution on isolated branches, reproducible results across 7 platforms, and features GitHub Agentic Workflows lacks like per-node model control.
DAG-Based Orchestration for Reproducible AI Agents
Define workflows as directed acyclic graphs (DAGs) in YAML, where each node is an AI prompt, command, script, loop, or human approval gate. Independent nodes run in parallel automatically, ensuring the same inputs always produce identical outputs—unlike raw agents' unpredictability. Every workflow uses an isolated Git worktree on a dedicated branch, enabling simultaneous runs (e.g., five features without merge conflicts) with auto branch creation and cleanup. Visual builder shows color-coded nodes (blue for prompts, green for commands) syncing to editable YAML, eliminating black-box debugging. Enterprise controls include hooks for intercepting tool calls, per-node tool whitelisting/blacklisting, model overrides (Haiku for classification, Opus for implementation), and MCP for external tools like GitHub or Postgres.
Issue-to-PR Pipeline with Parallel Reviews
Drop a GitHub issue link into Archon's chat to trigger: codebase analysis, implementation plan, code writing, validation, and PR creation. Five parallel review agents then assess code quality, error handling, test coverage, comment quality, documentation, and impact; a synthesizer consolidates reports, and an autofixer patches critical issues. Per Entropics 2026 report, structured summaries boost developer acceptance of AI changes from 62% to 89%. Ships with 17 ready workflows (issue fixing, smart review, refactoring) and 36 command templates, all customizable. Execution logs reveal every AI decision for full auditability.
Cross-Platform Access and Team Dashboard
Trigger workflows from web UI, CLI, Slack, Telegram, GitHub mentions, Discord, or Git—seven platforms sharing conversation history and codebase state, so you start on mobile and check PRs later. Dashboard overviews all runs by status (running, paused, failed), duration, and agent assignments; drill into glass-box execution traces.
Edges Over GitHub, Devon, and Raw Agents
Raw agents (Claude Code, Cursor) lack reproducibility, parallelism, and trails. GitHub Agentic Workflows are native but miss DAGs, parallel layers, model control, and gates. Devon is autonomous ($20-$500/month) but black-box. Archon is self-hosted (Docker Compose + SQLite, code stays local), paying only API costs. Replaces original 13k-star Archon (now redundant post-RAG natives in Claude/Codex). Launching soon on GitHub with livestream; join 1k+ builder community for daily hangouts, workshops, and 72-lesson course to avoid 91% solo dropout rate in 3 months.