Skip to content
little big brain

What is little big brain?

little big brain (lbb) is an RDF database that keeps its truth in object storage, written in Rust. You load RDF, query it with SPARQL 1.1, validate it with SHACL, and read any past state of it by pinning a snapshot.

The object-storage design explains the rest of the system. A bucket (S3-compatible or a local filesystem) is durable, cheap, and shared. If the database writes its truth and its indexes there as immutable objects, then the servers in front of it hold no durable state. They can be restarted, replaced, or scaled out, and every one of them reads the same bytes. What you operate is a bucket plus disposable compute.

What it is good at

Section titled “What it is good at”
  • Standards-based query. A conformant SPARQL 1.1 engine, served both as JSON endpoints and as the native SPARQL 1.1 Protocol at /sparql, so YASGUI, Protégé, and RDFLib connect without an adapter. A structured SELECT/aggregate surface covers the same graph for callers that prefer JSON over query text.
  • Validation and reasoning over the same data. SHACL Core, SHACL-SPARQL, and SHACL-AF validate shapes and entail new triples at publish time. Opt-in RDFS entailment lets a query for a broad class match its subclasses. Ontologies import from Turtle, JSON-LD, RDF/XML, and CSV/TSV, including OWL class and object-property vocabularies.
  • Immutable published snapshots. A read is served from one immutable generation identified by a sequence number. Every response says which sequence it answered from, so a result set stays reproducible.
  • Time travel and branches. Pin a query to a past commit with as_of_commit_seq, or fork a copy-on-write branch, load into it, validate it, and merge it back as one commit.
  • Object storage as the source of truth. The durable graph, its write-ahead log, and its index runs live in object storage. Indexes are derived acceleration structures that can be rebuilt and garbage-collected at any time.

Full-text and vector search are not documented yet.

How you talk to it

Section titled “How you talk to it”
Surface Best for
TypeScript SDK (@littlebigbrain/client) Node, browser, and edge apps
Python SDK (lbb) Scripts, notebooks, data workflows
MCP server (@littlebigbrain/mcp) Giving an AI agent graph-aware tools
HTTP API Any language, plus the native SPARQL 1.1 Protocol

All of the application SDKs speak the same HTTP API and authenticate with the same stack API key, so you can mix them freely: load with Python, query from TypeScript, let an agent read through MCP, and explore in the console, all against one graph.

Deployment shape

Section titled “Deployment shape”

Today little big brain is a hosted product. Each stack has a managed data endpoint at <tenant-short-id>--<stack-slug>.db.eu.littlebigbrain.com, with an account/stack API at api.littlebigbrain.com, a hosted MCP endpoint at mcp.littlebigbrain.com, and the console at cloud.littlebigbrain.com. You sign up, create a stack (an isolated workspace), and get a per-stack API key.

A self-hostable distribution is planned: a Docker container or pre-built server that runs against a local filesystem or an S3-compatible bucket.

Where to go next

Section titled “Where to go next”