decapod 0.1.4

🦀 Decapod is a Rust-built, repo-native control-plane kernel for AI swarms—safe shared state, enforced truth, and loop-agnostic orchestration.
Documentation

Quick Start

cargo install decapod
cd your-project
decapod init

That's it. decapod init creates (and backs up existing) CLAUDE.md, AGENTS.md, and GEMINI.md with methodology your agents follow.

What agents get:

  • Persistent memory across sessions
  • Standard interface (no guessing)
  • Proof requirements before claiming "done"

What you get:

  • Embedded constitution that all agents follow
  • Confidence to merge agent work
  • Simple overrides via OVERRIDE.md

The Problem

AI lowers the barrier to writing code—but shipping code is still hard.

Every agent session starts from scratch. Context evaporates. You can't trust it to:

  • Remember what it built yesterday
  • Follow your standards without drift
  • Prove the code works before claiming "done"
  • Ship without you checking every line

Shipping is a system. Decapod is that system for agentic development.

What It Does

You wouldn't run microservices without coordination. You wouldn't run a database without ACID. But we're handing agents write access to production repos and hoping "good prompting" substitutes for discipline.

Decapod turns agent output into an engineering pipeline:

  • Shared state that survives sessions — work doesn't reset on handoff
  • One agent-first interface (CLI + schemas) — agents don't poke internals
  • One authority chain — Intent → Spec → Code → Proof → Merge
  • Proof gates — "sounds right" can't ship without evidence
  • Full traceability — what changed, who changed it, why

Env Layout

.decapod/
├── data/        # state that survives sessions
└── OVERRIDE.md  # project-specific constitution overrides (optional)

Local-first by design:

  • No daemon
  • No hosted service
  • No "agent memory SaaS"
  • Just a repo-native control plane you version, review, and enforce

Who This Is For

✅ You're building real products with AI agents ✅ You want CI/CD discipline, not "vibes-based" shipping ✅ You need multiple agents working without chaos ✅ You merge to production (not just prototyping)

If you want a one-off script, Decapod is overkill. If you want agents to ship production code, Decapod is the missing layer.

Get Involved

  • Ship a subsystem — adapters, proof harnesses, connectors
  • Harden the constitution — if you found a rule that stops drift, PR it upstream
  • Break it — open issues with repros (they become proof gates)
  • Share overrides — Found a useful project pattern? Add it to OVERRIDE.md examples