🕸 Cross-Layer Accountability

The accountability layer for the autonomous-agent era. Mneme links code ↔ database ↔ API ↔ business rules into one deterministic graph — and asks the questions a single-layer tool can't. No LLM in the analysis path — every finding is reproducible and signed.

npm i -g mneme-ai && mneme review

mneme review · loading a live example…

cloning a real public repo + running the cross-layer suite, live…

The 10 checks

each one answers a question nothing else answered — and your AI agent gets them all as MCP tools, automatically

🔍

Codebase Review

the one-command report: grade + risk + authz + tests

mneme review

🛰

Impact Radar

see a change ripple across code · data · api · business

mneme graph view <fn>

⛔

Drop Safety

what breaks if you remove this table? SAFE / RISKY / CRITICAL

mneme graph reverse <table>

💥

Agent Collision

two agents/branches colliding across DIFFERENT files — git is blind to it

mneme collision --branches a,b

🤝

Scope Covenant

did the agent stay in the scope it declared? signed, cross-vendor

mneme scope verify

🏷

Commit Honesty

a 'fix typo' that secretly rewrites a payment keystone

mneme commit-check

🧪

Test Gap

the keystone (sole writer to a table) no test even mentions

mneme testgap

🎯

Risk Hotspots

every signal fused into one 'what to guard first' ranking

mneme risk

🔒

Authz Gap

an endpoint that writes a sensitive table with no auth on the path

mneme authz

🧭

Onboarding Path

the real data-flows to read first — orient in a new repo fast

mneme onboard

📊 Measured extractor accuracy — precision 100% · macro-F1 0.99 across 8 dimensions on a labeled corpus (it proves the accuracy, doesn't claim it) — audit /api/accuracy

Deterministic · signed · local-first · works on JS/TS · Python · Go · Rust · Ruby · Java · C# · the source never leaves your machine (the demo clones to a temp dir, scans, and deletes).
npm · /review · /radar · _{honest: each finding is a candidate to inspect, not a proof of a runtime bug.}