<p align="center">🦀</p>
<p align="center">
<code>cargo install decapod && decapod init</code>
</p>
<p align="center">
<strong>Decapod</strong><br />
A daemonless control plane for AI coding agents.
</p>
<p align="center">
Called on demand inside agent loops. No background process, no new workflow, local-first state you can verify.
</p>
<p align="center">
<a href="https://github.com/DecapodLabs/decapod/actions"><img alt="CI" src="https://github.com/DecapodLabs/decapod/actions/workflows/ci.yml/badge.svg"></a>
<a href="https://crates.io/crates/decapod"><img alt="crates.io" src="https://img.shields.io/crates/v/decapod.svg"></a>
<a href="https://github.com/DecapodLabs/decapod/blob/master/LICENSE"><img alt="License: MIT" src="https://img.shields.io/badge/license-MIT-blue.svg"></a>
<a href="https://ko-fi.com/decapodlabs"><img alt="Ko-fi" src="https://img.shields.io/badge/Support-Ko--fi-ff5f5f?logo=ko-fi&logoColor=white"></a>
</p>
---
## Why Decapod 🧠
AI coding agents are strong at generating code. Most failures happen before and after generation: unclear intent, fuzzy boundaries, and weak completion checks.
Decapod is the missing layer in that loop. Agents call it mid-run to lock intent, enforce boundaries, and prove completion with explicit gates. It shapes inference without doing inference.
Decapod is **daemonless**. There is no long-lived service. The binary starts when an agent calls it and exits immediately after the call.
"Just use Decapod" is literal:
- `cargo install decapod`
- `decapod init`
Then continue with Claude Code, OpenAI Codex, Gemini CLI, Cursor, or any tool that can invoke a CLI command. Decapod is agent-agnostic and safe for concurrent multi-agent execution.
State is local and durable in `.decapod/`: shared context, decisions, and traces persist across sessions and remain retrievable over time.
Related: [Evaluating AGENTS.md](https://arxiv.org/pdf/2602.11988) (ETH SRI, 2026) on context-file quality and agent cost/performance.
<p align="center">
☕ Like Decapod? <a href="https://ko-fi.com/decapodlabs"><strong>Buy us a coffee on Ko-fi</strong></a> 💙
</p>
## Getting Started 🚀
```
cargo install decapod
decapod init
```
Then keep using your agents normally. Decapod is called from inside those agent runs when control-plane decisions are needed.
Agent integration: If you use Claude Code / Codex / Gemini / Cursor / similar tools, see `AGENTS.md` and the tool-specific entrypoint files (`CLAUDE.md`, `CODEX.md`, `GEMINI.md`) for the exact operational contract.
Learn more about the embedded [constitution](constitution/core/DECAPOD.md).
Override constitution defaults with plain English in `.decapod/OVERRIDE.md` after you initilaize Decapod in your project directory.
## Assurance Model ✅
Decapod centers execution around three outcomes:
- `Advisory`: clear next actions that tighten intent and reduce wasted loops.
- `Interlock`: hard policy boundaries that block unsafe or out-of-contract flow.
- `Attestation`: durable, structured proof that completion criteria actually passed.
## Operating Model ⚙️
```text
Human Intent
|
v
AI Agent(s) <----> Decapod Runtime <----> Repository + Policy
| | |
| | +-- Interlock (enforced boundaries)
| +------- Advisory (guided execution)
+------------ Attestation (verifiable outcomes)
```
## Features ✨
- ✅ Daemonless execution: no background agent manager, no hidden runtime.
- ✅ Two-command adoption: `cargo install decapod` and `decapod init`.
- ✅ Agent-agnostic contract: one CLI/RPC surface across Claude, Codex, Gemini, Cursor, and others.
- ✅ Parallel-safe collaboration: multiple agents can operate in one repo without state collisions.
- ✅ Work Unit governance: `intent/spec/state/proof` is explicit, durable, and machine-checkable.
- ✅ VERIFIED is enforced: completion requires passing proof-plan results, not narrative claims.
- ✅ Promotion discipline: validate-time and publish-time gates block silent regressions.
- ✅ Deterministic context capsules: scoped (`core|interfaces|plugins`), hashable, and reproducible.
- ✅ Variance-aware eval kernel: repeat-run plans, strict judge contracts, statistical regression gates.
- ✅ Knowledge promotion firewall: procedural truth requires event-backed provenance in `.decapod/data/knowledge.promotions.jsonl`.
- ✅ Shared transient aptitude memory: capture human-taught preferences once, reuse across agents and sessions.
- ✅ Plain-English policy control in `.decapod/OVERRIDE.md`.
- ✅ Local-first auditability: `.decapod/` keeps durable traces, decisions, and proof artifacts.
And dozens more. For the full high-level and data-level surface area, see `decapod docs show core/INTERFACES.md` and the override template at `.decapod/OVERRIDE.md`.
## Contributing 🤝
```bash
git clone https://github.com/DecapodLabs/decapod
cd decapod
cargo build
cargo test
decapod validate
```
## Documentation 📚
- Development guide: [CONTRIBUTING.md](CONTRIBUTING.md)
- Security policy: [SECURITY.md](SECURITY.md)
- Release history: [CHANGELOG.md](CHANGELOG.md)
## Support 💖
- 🐛 [File an issue](https://github.com/DecapodLabs/decapod/issues)
- ☕ [Support on Ko-fi](https://ko-fi.com/decapodlabs)
## License 📄
MIT. See [LICENSE](LICENSE).