Bring your own repo
Rewind works with any GitHub repository. For this hackathon, the demo memory is built from the Cognee repo's own issues and pull requests.
Memory graph
Every entity and relationship Cognee extracted, rendered as a live force graph. Watch your agent's memory take shape.
Retrieval X-ray
Ask a question and see exactly which nodes and edges produced the answer. The retrieved subgraph glows, everything else fades.
Time travel
Scrub back through ingest batches to see what the agent knew at any point, and ask questions against that older memory.
Node inspector
Click any node for its type, properties, source document, ingest batch, and connections. No more opaque memory.
FAQ
Questions worth asking
What is Rewind?+
Rewind is a memory debugger for AI agents built on Cognee. It renders the agent's knowledge graph, shows exactly which memories produced each answer, and lets you rewind memory to an earlier point in time.
What data is in the demo?+
The first 45 issues and pull requests of the Cognee repository, ingested through Cognee itself in three chronological batches: 292 nodes and 859 relationships extracted by an LLM.
Does it only work with the Cognee repo?+
No. The ingestion pipeline is generic: point it at any GitHub repository and it turns the issues and pull requests into a memory graph. We focused this demo on the Cognee repo because the hackathon is judged by its maintainers. Broader repo support in the hosted version is on the roadmap if there is demand.
How does the retrieval X-ray work?+
When you ask a question, Cognee returns the exact triplets it retrieved to answer it. Rewind extracts their node IDs and highlights precisely those nodes and edges in the graph, so the provenance is real, not approximated.
How does time travel work?+
Each ingest batch is tagged as a native Cognee NodeSet, so batch membership lives in the graph itself. Dragging the timeline restricts retrieval to the batches that existed at that point, meaning answers come only from what the agent knew then.
Is it open source?+
Yes. The full stack (FastAPI + Cognee backend, Next.js frontend) is MIT-licensed on GitHub.