What is little big brain?
little big brain is a search platform for AI applications and agents. It puts three things in one system, over one copy of your data:
- Search by meaning. Vector search finds records that match a question in other words. We run the embedding models and keep the vectors current.
- A knowledge graph. Records are linked. A graph query follows the links from a search hit to the facts behind an answer.
- History. Each write is a commit. Each response names the commit it read, so you can replay an answer later.
Why search and graph belong together
Section titled “Why search and graph belong together”Vector search has high recall. It finds text that means the same thing in other words. It cannot show how two facts connect, or whether an answer is complete. An agent then reads more chunks and guesses the connections.
A graph query is precise. It returns only what the stored links support. It needs a starting point, and a question in natural language rarely names one.
little big brain uses both, at one commit. Search finds the starting point. The graph proves the answer. The response says which version of the data it used.
What you build with it
Section titled “What you build with it”- Agents that cite their answers. An answer comes with the records, links and evidence behind it.
- Support and operations assistants. “Which release fixed the login bug?” The quickstart builds this one.
- Tracing and exposure questions. Which batches contain a recalled lot? Which services depend on a vulnerable library? See use cases.
- Audit trails. Replay the query behind a decision on the data the agent read then.
What the platform runs for you
Section titled “What the platform runs for you”- Embeddings. The vectors live next to the records they describe.
- Re-embedding. A record is embedded again when its fields change.
- Agreement between search and graph. Search hits are checked against the graph at one commit.
- Storage and servers. Your graph lives in object storage. Servers cache hot data on NVMe and hold no durable state. You pay per GiB stored.
How you use it
Section titled “How you use it”| Surface | Best for |
|---|---|
TypeScript SDK (@littlebigbrain/client) |
Node.js, API routes and edge functions |
Python SDK (pip install littlebigbrain) |
Scripts, notebooks and data pipelines |
MCP server (@littlebigbrain/mcp) |
Giving an agent search, query and write tools |
| HTTP API | Any language, and standard SPARQL tools |
All clients call the same HTTP API with the same stack API key. Load data with Python, query from TypeScript and let an agent read through MCP. All of them work on one graph.
Built on open standards
Section titled “Built on open standards”You write records as JSON. Underneath, they are stored as RDF, the W3C standard for linked data. Queries use SPARQL 1.1, validation rules use SHACL, and reasoning uses RDFS and OWL.
You do not need to know these standards to start. They mean that standard tools such as YASGUI, Protégé and RDFLib connect directly. If your data is RDF already, you can load it as it is. The core concepts explain the words you will meet.
Where it runs
Section titled “Where it runs”little big brain is a hosted service in the EU. Each stack has its own endpoint
at <tenant-short-id>--<stack-slug>.db.eu.littlebigbrain.com. You sign up in
the console, create a stack and copy its
API key. The hosted MCP server is at mcp.littlebigbrain.com.
A self-hosted distribution is planned. It will run against a local filesystem or an S3-compatible bucket.
Where to go next
Section titled “Where to go next”- Run the quickstart: search, follow a path and cite a commit.
- Learn the model in core concepts.
- See how the steps fit together in how search works.