Keeping architecture consistent across multiple AI agents.
Three actors touch the same service in three separate sessions, sharing no memory with each other. They share one thing: the compiled decision corpus. The invariants hold because they live outside every actor, not inside any of them.
ADR-001 forbids external databases. Redis is forbidden.ADR-003 forbids ORMs in v1.The demo shows how three independent actors stay aligned because they all evaluate against this same compiled corpus — not against each other.
What happened
- Actor A proposed Redis. ADR-001 returned
FAIL; a second, corrected file checked independently and passed. - Actor B, in a fresh session with no memory of A, built on the corrected primitive and passed.
- Actor C's remediation reached for an ORM. ADR-003 returned
FAIL; the compliant remediation passed.
Run it yourself
git clone https://github.com/MnemeHQ/mneme
cd mneme/examples/multi-agent-governance
python run.pyActor C is the interesting one — a genuine architectural violation surfaced explicitly rather than silently resolved:
FAIL [ADR-003] anti_pattern "sqlalchemy" -- trigger: sqlalchemy
WARN [ADR-003] constraint "no sqlalchemy" -- trigger: sqlalchemy
Result: FAILThe actors are scripted so the enforcement output stays deterministic. This is not a multi-agent runtime and does not claim to be one — it shows that one corpus keeps three independent sessions coherent.
Ready to try it?
Install the CLI and check your first decision in under a minute.