Free AI Coding Agent — Open Source

Stoneforge is a free AI coding agent that orchestrates multiple AI agents working on your codebase in parallel. As an open source AI coding assistant licensed under Apache 2.0, there are no per-seat fees, no usage caps, and no premium tiers.

You bring your own LLM API keys (Anthropic, OpenAI, or others), and this free AI coding assistant handles the orchestration: dispatching tasks to agents, managing isolated git worktrees, and merging results back into your codebase.

What you get for $0

This open source AI coding agent ships with the full orchestration stack:

• Multi-agent dispatch. A Director agent breaks down work into tasks. A daemon assigns those tasks to Worker agents running in isolated git worktrees. Multiple agents work on separate branches simultaneously.
• Automated merge and review. Steward agents review completed work, run tests, resolve straightforward merge conflicts, and merge branches. See automated code review for details.
• Web dashboard. A local web UI at localhost:3457 shows agent status, task progress, plans, and messages in real time.
• Dependency-aware scheduling. Define task dependencies so the orchestrator sequences work correctly. Independent tasks run in parallel; dependent tasks wait for their prerequisites.
• Knowledge persistence. Workspace documents survive across sessions. Agents build up a shared knowledge base about your codebase over time.

None of these features are gated behind a paid plan.

No API key middleman

Stoneforge does not proxy your LLM calls or add markup to token costs. You configure your API keys directly, and agents call providers at their standard rates. If Anthropic charges $3 per million input tokens for Claude Sonnet, that is what you pay. Stoneforge takes no cut.

This also means you are not locked into one provider. You can use Claude Code for some agents and OpenAI Codex for others within the same workspace.

Your code stays on your machine

This free AI coding agent runs locally. Your source code, task history, and orchestration state live on your filesystem in SQLite databases and git-tracked JSONL files. Code only leaves your machine when an agent sends context to an LLM API, and you control which provider that is.

For air-gapped environments, this open source AI coding agent works with local models. Nothing needs to phone home.

How it compares to paid alternatives

The AI coding tool market has grown expensive:

• Devin starts at $20/month for individuals, $500/month for teams, plus per-unit charges for additional compute.
• Cursor Pro costs $20/month per seat, with credit-based billing for premium model access.
• Claude Code with Team plans runs $25-$150 per seat per month depending on usage tier.

Stoneforge costs nothing to run. The trade-off: you self-host it and manage the setup yourself. There is no managed cloud version or support team. You are responsible for your own infrastructure.

For solo developers and small teams comfortable with self-hosting, a free AI coding assistant is a straightforward win. For organizations that need managed hosting and vendor support, a paid tool might be a better fit.

Get running in minutes

# Install globally
npm install -g @stoneforge/smithy

# Initialize in your project
cd your-project
sf init

# Start the orchestration server
sf serve

Open http://localhost:3457. Register a Director agent, add your LLM API keys, and create your first plan. The Director breaks the plan into tasks, the daemon dispatches them to Worker agents, and Steward agents handle review and merge.

The quick start guide walks through the full setup, including agent configuration and your first multi-agent run. You can also explore parallel AI development to see how the orchestration works in practice.