CockroachDB x AWS Hackathon

CommonMind shared memory.

A shared memory layer for humans and agents: capture what happened, recall it semantically, stop for human approval when the stakes matter, and preserve the decision as memory for the next run.

Open CommonMind →
CockroachDB CloudC-SPANN Vector IndexingAmazon BedrockAWS LambdaSNS / SQS / S3TypeScript / React
The Challenge

Make agent memory durable

Agents can learn a codebase, make progress, and still start the next session cold. The deeper problem is not just lost context; it is the missing shared memory between agents and the humans who need to understand, approve, and reuse their work.


CommonMind was built for the CockroachDB x AWS Hackathon to treat memory as the product rather than a bolt-on vector store. The Devpost submission is available here: CommonMind Shared Memory for Humans and Agents.

What Shipped
  • Atomic Capture A memory record and its embedding commit together, so there is no gap between what happened and what can be retrieved.
  • Vector Recall Semantic recall runs inside CockroachDB using C-SPANN over VECTOR(1024), keeping retrieval close to the system of record.
  • Human Approval Agents can pause before consequential action, ask a human for a decision, and store that decision as reusable memory.
  • Multiple Proof Surfaces The same memory core supports trading workflows, coding sessions, company memory, and an interactive remembered-experience demo.
The Solution

One Source of Truth

CockroachDB stores the memory log, embeddings, approvals, and event source, avoiding split-brain behavior between a database, queue, and vector store.

Serverless Agent Path

AWS Lambda, Amazon Bedrock, SNS, SQS, and S3 keep the execution layer elastic while preserving CockroachDB as the durable memory core.

Reliability as Product Behavior

The architecture favors strong consistency and fail-closed behavior, because stale memory is more dangerous than an honest unavailable state.

Results
1 txn
Record + Embedding
4
Proof Surfaces
3x
Default Replication

BUILDING AN AGENT MEMORY LAYER?

We design AI systems that remember decisions, preserve context, and keep humans in control of consequential work.

Start the Conversation →