OpenAgents is a local-first stack for autonomous agents that can sign, run tools, verify work, account for cost, and (optionally) buy compute in open markets. The wedge product in this repo is Autopilot: an autonomous coding agent that plans → executes → verifies → emits replayable artifacts.
If you're looking for the philosophy / "why open", start with MANIFESTO.md.
git clone https://github.com/OpenAgentsInc/openagents.git
cd openagents
# build + run the Autopilot app
cargo build -p autopilot
cargo run -p autopilotRun a one-off task via CLI mode:
cargo run -p autopilot -- run "Summarize @README.md and run !git status"Tip: this repo defines a Cargo alias in .cargo/config.toml, so cargo autopilot ...
is shorthand for cargo run -p autopilot -- ....
Autopilot is designed to be:
- Local-first: runs against your repo on your machine by default
- Verifiable: uses deterministic checks (tests/builds) as the ground-truth loop
- Inspectable: emits structured session logs and artifacts for replay/debug/audit
- Optimizable: uses DSPy-style signatures/modules/optimizers to improve policies over time
Start here depending on what you're doing:
- Terms / vocabulary: GLOSSARY.md
- Run + ship priorities: ROADMAP.md
- Architecture + strategy: SYNTHESIS.md
- Current implementation reality: SYNTHESIS_EXECUTION.md
- Formal write-up: PAPER.md
- Repo layout + crate map: PROJECT_OVERVIEW.md
Autopilot-specific docs:
- crates/autopilot/docs/MVP.md
- crates/autopilot/docs/ROADMAP.md
- crates/autopilot-core/docs/EXECUTION_FLOW.md
DSPy / dsrs docs:
If you're trying to find "where does X live?", use PROJECT_OVERVIEW.md. It's the repo map (crates, responsibilities, data flows).