Roadmap
What we're building next.
Streamlined Setup & Binary Install
Getting started with Archon should take under 5 minutes. A fully self-contained binary, one-line installers, and a first-run wizard that auto-detects your AI provider.
- Fully self-contained binary distribution (no Node/Bun required)
- One-line install: curl/irm scripts for macOS, Linux, Windows
- Homebrew formula
- Interactive first-run setup wizard
- Auto-detection of Claude and Codex credentials
Workflow Marketplace
An open source registry for community-built Archon workflows. Browse, install, and share workflows in one place.
- archon.diy/workflows — searchable, filterable directory
- WORKFLOW.md spec — standard format for shareable workflows
- One-command install: archon workflow install <slug>
- Category filtering, tag search, and author pages
- PR-based submission — curated, no backend required
Eval System
A built-in evaluation framework to measure and improve workflow quality — test cases, correctness scoring, and reliability testing.
- WORKFLOW.eval.yaml — define test inputs and expected outputs inline
- Step-level and output-level correctness scoring
- Reliability testing across multiple runs
- Eval score badges on marketplace listings
- archon workflow eval <name> — run evals from the CLI
Advanced Workflow Control Flow
Make workflows expressive enough for real test/fix and approval-driven automation — multi-node loop bodies, branching on approvals, and a real expression evaluator.
- Multi-node loop bodies (compose plan → implement → validate per iteration)
- Conditional branching on approval and reject outcomes
- Semantic completion signals — agents finish when done, not by token-matching
- Real expression evaluator powering when:, loop_until:, and condition: clauses
Persistent Project Orchestrator
One stateful conversation per codebase that retains context across runs, with project-first navigation and observable subagent activity.
- Persistent orchestrator session bound to each project
- Project memory carried across runs and restarts
- Projects-first navigation in the Web UI
- Live SDK lifecycle events (subagent + hook activity) streamed to the UI
Local LLM Support
Connect Archon to local inference servers and OpenAI-compatible API proxies — bring your own model.
- OpenAI-compatible baseURL configuration
- Runtime workflow variables for dispatch-time model selection
- Bundled workflows respect DEFAULT_AI_ASSISTANT instead of locking a provider
Workflow Execution Reliability
Eliminate silent-success failure modes — workflows must never report completion while shipping nothing.
- Audit and harden every workflow resumption and cache path
- Defensive handling of provider error patterns (Codex turn.failed, Claude stop-sequence success)
- Invariant checks before state restore — refuse to auto-resume failed runs into fresh requests
◇ Future
Additional Model Providers
Pluggable provider SDKs beyond Claude and Codex — Copilot, Hermes — plus per-dispatch model selection.
Multi-Repo Workspace Support
Multiple clones of the same remote as distinct projects, branch-aware sync, and unambiguous webhook routing.
Enterprise GitHub Auth
GitHub App with per-installation tokens and secure secret resolution for org and team setups.
✓ Shipped
Core CLI & DAG Engine
The foundation — YAML-defined workflows, DAG execution with dependency resolution, and git worktree isolation per run.
- YAML workflow definition language
- DAG orchestration with full dependency resolution
- Git worktree isolation — no conflicts between runs
- Multi-provider support (Claude Code SDK, Codex SDK)
Adapters & Deployment
Trigger Archon workflows from any surface — Slack, Telegram, GitHub, Web UI, Discord — and deploy anywhere.
- Official adapters: Slack, Telegram, GitHub, Web
- Community adapters: Discord, GitLab, Gitea
- Docker & cloud deployment guides
- Windows native support
Pi Agent Provider
Run Archon workflows with Pi — an open-source, multi-model coding agent. Breaks the Claude Code / Codex lock-in and opens Archon to GPT, Gemini, Qwen, DeepSeek, and more.
- Pi coding agent as a first-class Archon provider
- Supports any model Pi runs on — GPT-5, Claude, Gemini 2.x, Qwen Coder, DeepSeek V3
- Optional access to Pi's ~540-package extension ecosystem
- Skills, structured output, session resume, and tool restrictions all supported
Hooks, Commands & Quality Gates
Fine-grained workflow control with hooks, reusable commands, loop nodes, approval gates, and script nodes.
- Pre/post hooks on any workflow node
- Reusable commands library (share logic across workflows)
- Loop nodes with configurable iterations
- Approval nodes for human-in-the-loop checkpoints
- Script nodes for custom automation logic