Project Name
Fleet Consensus Ledger (FCL)
Inspiration
Most multi-agent AI systems rely on fragile local state management. When an agent crashes mid-task, data gets silently corrupted or workflows get stuck forever. Medusa Black Labs built FCL to solve the data consistency and resilience challenges inherent in large-scale autonomous agent coordination.
What it does
FCL coordinates a fleet of AI agents working asynchronously on a shared mission. Using CockroachDB as a globally distributed transactional memory layer, agent writes stay consistent even under concurrent load. If a running agent fails or drops mid-task, its task lease automatically expires — a sibling agent detects the expired lease, reclaims the task, and finishes it with no human intervention.
How we built it
The core engine runs on AWS Lambda (serverless, arm64). We implemented an application-level retry loop with exponential backoff and random jitter to handle CockroachDB's strict SERIALIZABLE isolation conflicts (SQLSTATE 40001). Every state write is signed with a SHA-256 hash recording the acting agent's identity and access level, giving a full audit trail.
Built With
- aws-cloudwatch
- aws-eventbridge
- aws-lambda
- aws-sam
- aws-secrets-manager
- cockroachdb
- mcp
- psycopg
- python
Log in or sign up for Devpost to join the conversation.