Agentic AI governance in operation

Watch an agent cross the line. Then watch Mneme stop it.

Two runnable demos show architecture failures before they land. Then see Mneme block a violating write in a real Claude Agent SDK workflow before it reaches disk.

Runnable proofs

The change looks useful. The decision says otherwise.

Core enforcement Local CLI · DENY → PASS

A second database sneaks in.

JSON becomes Postgres. Mneme stops the migration, then checks the safer retry.

If it lands One product. Two sources of truth.

See how Mneme blocks it →

Drift prevention Retrieval architecture · Local CLI · DENY → PASS

The rulebook starts guessing.

A vector-search change lets agents find different decisions for the same task. Mneme blocks the fork.

If it lands Same request. Different rules. Conflicting code.

See how Mneme blocks it →

Claude Agent SDK Native integration · Live + deterministic · DENY → PASS

Claude proposes a write. Mneme stops it before disk.

A real Claude Agent SDK workflow receives the governing decision, attempts a violating Write, gets denied, and retries with the compliant change.

Run the Agent SDK proof →

01Claude proposesWrite → storage.py

02Mneme deniesDecision injected · hook returns DENY

03Claude retriesApproved path → PASS

Try it yourself

Run a governed check locally.

Open source. Deterministic. No model call required.

pipx install "mneme-hq>=0.9.0"

Supporting examples

More ways Mneme enforces architecture.

CI boundaries, dependency policy, ADR import, and compliant checks.

CI enforcement · Strict mode · DENYThe code works. The boundary does not.

Stop direct database access in a service.

Migration scope · Strict mode · DENYThe pattern is everywhere. Here it is forbidden.

Scope a decision to the migrated tree only.

Dependency policy · Warn mode · WARNNew package? Make it a decision.

Flag the dependency before it drifts in.

ADR import · Local CLI · WARNTurn ADRs into checks.

Compile the rules you already maintain.

Compliant check · Local CLI · PASSKeep JSON. Ship the change.

Verify a clean storage task.

What comes next

Multi-agent governance is the target architecture.

Multi-agent governance Target architecture · Not shipped · Not runnable

Many agents. One rulebook.

The goal: every agent gets the same rule, meets the same check, and leaves one shared record.

Inspect the target contract →

Same rule version Agent + result identity Connected retry record

An engineer reviewing architecture and product flows drawn on a whiteboard — Architecture begins as a plan people can inspect. Agentic governance keeps later changes tied to it.