today
- half your AI bill goes to re-explaining the repo
- every chat starts from zero context
- the refactor that failed last week gets tried again
- no idea which model is good at which task type
// governance + audit layer for AI coding agents
[ repo-bound memory ] · build v4.6.1
one rulebook across every CLI · every 0dai run --now leaves a receipt · your source stays local
Production-grade AI agent governance
Govern your Claude CodeCodexOpenCodeGeminiAiderQoderKimi Code agents with a 19-principle production doctrine.Open-core. Self-hostable.
~/repo $ live install
$ npm i -g @0dai-dev/cli
fetching @0dai-dev/cli@v4.6.1
$ 0dai init --local
detecting agents: claude-code, codex, gemini, aider, opencode, qoder
wrote CLAUDE.md · AGENTS.md · ai/manifest/ — no account needed
$ 0dai activate free
free account (no card) · adds the full per-CLI config set + sync + run
terminal CLIs
registered MCP tools
two ways to run agents.
only one of them compounds.
I built 0dai because I was tired of explaining my repo for the 47th time. The agents were not dumb — they were amnesiac. So I made the memory layer they should have had. Solo founder, shipping in public, dogfooded on 0dai's own repo.
with 0dai
- context delivered, not re-paid for
- projects remember decisions, outcomes, patterns
- failed paths flagged before agents repeat them
- model routing based on what actually shipped
Not a fit for every team. See where Claude Code subagents beat 0dai — the honest feature matrix.
deliberation becomes a product surface.
Memory load, routing, consensus, patch hold, QA proof, and outcome capture — made visible, not implied.
free stays free.
upgrade when you want the network.
Cancel Pro and you keep what is already in your repo. The cloud layer goes dark. No bait, no decay of local code.
Free (52 core MCP tools) · Pro $15/mo · Team $49/mo — see plans and the calculator →
from clean repo to first shipped task.
0dai init --local writes the scaffold, no signup. A free, no-card account adds sync and run.