Case Study / Public Source Available Reference Implementation / Enterprise Validated
Multi-Agent Memory Coordination System
A deployable private-intranet, source-available reference implementation for coordinating external AI agents through registered identities and scoped grants, persistent reviewed memory, meeting and current-message routing, acknowledgements, and hash/CAS/lease controls for simultaneous local .uai edits.
External AI agents working on the same initiative can lose durable context, duplicate work, overwrite local continuity files, or act outside their intended scope when identity, routing, memory, and edit coordination remain implicit.
Legacy risk
What could regress
The service must coordinate external agents and shared memory on a private intranet without becoming the agent runtime, silently promoting unreviewed messages, or granting authority beyond explicit scopes.
Validation method
How the claim is checked
Review the public source and companion documentation, run the repository validation suite, and trace identity, scope, memory, message, acknowledgement, hash/CAS, claim, lease, and persistence contracts while checking every explicit non-claim.
Architecture strategy
How the proof is structured.
Architecture evidence is presented with private implementation details abstracted and explicit not-claimed boundaries.
Make agent identity and scope explicit before accepting shared coordination or memory operations.Separate transient current messages from durable memory that has passed a public-safe review boundary.Route meetings, goals, tasks, targeted messages, broadcasts, and acknowledgements through reviewable contracts.Use hashes, compare-and-swap preconditions, bounded edit claims, and 60-1800 second leases to coordinate simultaneous local .uai edits without claiming automatic merge or distributed consensus.Keep the coordination service independent from model hosting or execution and require human verification where authority matters.
Implementation evidence
The public repository implements agent registration, immutable scope grants, persistent scoped memory, meeting-room coordination, current-message routing, and per-recipient acknowledgements.
The local AI handoff coordination contract exposes hash-only heads, compare-and-swap checks, bounded edit claims, and 60-1800 second leases.
Persistence implementations cover file storage, SQLite, MySQL, and MariaDB while documentation keeps human verification and public-safe memory review explicit.
Repository review confirms the service coordinates external agents and does not host or execute their models.
Result / current status
A deployable, reviewable coordination and persistent-memory implementation is publicly inspectable; no hosted-service or production-scale result is asserted.
Technologies
.NET 9C#HTTP APIsJSONSQLiteMySQLMariaDBHash and CAS CoordinationBounded LeasesUAI Memory
Technical value
Multi-agent coordination architecture for external agents
Persistent AI memory separated from transient current-message delivery
Registered identity and immutable scope-grant design
Meeting, task, targeted-message, broadcast, and acknowledgement routing
Conflict-aware simultaneous local AI handoff edit coordination through hashes, compare-and-swap checks, bounded claims, and leases
Enterprise .NET implementation discipline with pluggable persistence and human verification boundaries
Not claimed / boundary
Source-available does not mean OSI-approved open source.
No hosted SaaS or production-scale operating history is claimed.
The service coordinates external agents; it does not host or execute models.
Hash/CAS checks and leases are not distributed consensus, CRDT semantics, exactly-once delivery, or operating-system file locks.
Simultaneous local AI handoff changes are not auto-merged, and current messages are not automatically promoted into persistent reviewed memory.
