little big brain documentation

little big brain is a search platform for AI applications. Vector search finds facts by meaning. A graph query follows them to the evidence. Each answer names the commit it read, so you can replay it.

You write records as JSON. You need no RDF or graph database experience to start.

QuickstartWrite a few records, find one by meaning and follow its links to a cited answer. About ten minutes.

Load your dataLoad a database and your documents as JSON, in four calls. Examples for five use cases.

Connect an agentGive an agent tools to search, query and write through MCP.

Search by meaningDeclare embeddings on your types and filter hits by type and by link.

From a question to a cited answer

Step	What it does	Go deeper
Find it by meaning	Vector search finds a starting record, even when the question uses other words.	Search by meaning
Prove it by path	A graph query follows the stored links from that record to the facts behind the answer.	How search works
Keep the evidence	The response names its commit. Save it with the answer and replay the query later.	Replay an answer

The quickstart runs all three steps on one example: which release fixed the login bug?

Start from what you know

• Vector stores. Collections, documents and metadata map to graphs, records and fields. See core concepts.
• Graph databases. SPARQL patterns read like Cypher MATCH clauses. See the Cypher mapping.
• SQL. Tables and joins map to types and links. See coming from SQL.
• RDF. Load Turtle or N-Triples as it is. See load RDF data.

What you get

Managed embeddings. Choose the fields to embed. The platform embeds each record and keeps its vector current. Search by meaning.

Checked writes. Declare rules for your records. A write that breaks a rule is refused with a report of what to fix. Validate writes.

Replayable answers. Past commits stay readable. Run a query against the data an agent read last week. Replay an answer.

Open standards. Queries use SPARQL, rules use SHACL, and data is RDF underneath. Standard tools connect directly. Core concepts.

Connect your application

TypeScriptNode.js, API routes and edge functions.

PythonScripts, services and notebooks.

MCP for agentsSearch, query and write tools for any MCP client.

HTTP & SPARQLAny language, or an existing SPARQL client.

Guides

• Load your data: a database and your documents, in five use cases.
• Add links with an LLM: join your database records and your document sections.
• Build with JSON records: a schema, checked writes, a chain of links and a replayed decision.
• Use cases: six agent decisions and the query behind each one.
• Query with SPARQL and validate with SHACL: filters, aggregates, paths and rules.
• Reasoning & inference: answers that use your type hierarchy.
• Ontology starters (experimental): shared types for CRM, documents and work management.

For setup, see API keys and stack endpoints. For costs, see usage and billing or the pricing calculator.