Shared context for AI agents

One API call.Every agent in your stackremembers everything.

A self-enriching knowledge graph that lets Claude Code, OpenCode, Cursor, and Hermes share context — typed links, graph intelligence, and autonomous enrichment. Polyglot storage: Postgres + Neo4j. MCP-native. 23 tools.

Search
7-stage hybrid pipeline
FTSpgvectorRRF
Graph Intelligence
Neo4j + GDS plugins
PageRankLouvainshortest pathsimilarity
Every interface
SDKnpm
|
CLIterminal
|
MCP23 tools
The gap

Every AI agent
starts from zero

You run five agents: Claude Code builds features, OpenCode reviews PRs, Cursor edits frontend, Hermes handles ops. None of them know what the others built, decided, or learned.

The result: you're the copy-paste layer. Repeating context. Re-explaining decisions. Watching agents rediscover things another agent already figured out yesterday.

One agent per task
Each agent is an island. No shared context. No awareness of what's already been solved.
Vector DB "memory"
Dumping embeddings into a vector store gives you retrieval — not understanding. No typed relationships. No graph structure.
Brainbase
Every agent reads from and writes to the same knowledge graph. Typed links connect people, projects, and decisions. Graph intelligence surfaces what matters. No manual sync.
Access

Three interfaces.
One brain.

Same knowledge graph, delivered the way you work — npm package, terminal command, or native MCP tool.

SDK

npm install brainbase-sdk

import { Brainbase } from "brainbase-sdk";

const brain = new Brainbase({
  apiKey: "bb_live_..."
});

// Search + graph traversal
const results = await brain
  .search("Garry Tan");
const page = await brain
  .getPage("people/garry-tan");

// Graph intelligence (Neo4j)
const ranks = await brain
  .pageRank(25);
const path = await brain
  .shortestPath("people/a", "people/b");

CLI

Terminal-native

$ npm install -g brainbase-cli
$ brainbase config set apiKey bb_live_...

$ brainbase search "Garry Tan"
$ brainbase health
$ brainbase page people/garry-tan
$ brainbase links people/garry-tan
$ brainbase graph

# Graph intelligence (Neo4j)
$ brainbase pagerank --limit 25
$ brainbase communities
$ brainbase shortest-path a b
$ brainbase similar people/garry-tan

MCP

23 JSON-RPC tools

// Drop into Claude Code, Cursor,
// OpenCode, or Hermes
{
  "mcpServers": {
    "brainbase": {
      "type": "http",
      "url": "https://brainbase
        .belweave.ai/api/mcp",
      "headers": {
        "Authorization":
          "Bearer bb_live_..."
      }
    }
  }
}

Read: search, query, get_page,
  get_links, get_backlinks,
  get_timeline, get_health,
  get_stats, get_graph,
  list_pages, traverse_graph

Write: put_page, delete_page,
  add_link, remove_link,
  add_timeline_entry

Graph: pagerank, communities,
  shortest_path, similar_pages

Triggers: upsert_trigger,
  list_triggers, run_triggers
Architecture

Polyglot storage.
Postgres for records. Neo4j for intelligence.

Clients
AI Agents (MCP)Web UICLI (brainbase)SDK (npm)
API
MCP (23 tools)RESTCronPublic /b/···
Storage
System of Record
Supabase Postgrespgvectorpageslinkschunkstimeline
Graph Projection
Neo4jGDS PageRankGDS Louvainsimilarity
Dream
extractembedorphanspatternsentity tiersgraph sync

Nightly autonomous cycle. Extracts wikilinks + timelines, generates embeddings, reconnects orphans, escalates entities, syncs to Neo4j graph projection. Runs without any human input.

Capabilities

Built for multi-agent teams

01

Hybrid search

A 7-stage gated pipeline: full-text search → pgvector similarity → reciprocal rank fusion → compiled truth boost → backlink boost → intent-aware re-ranking → structured query handlers. Agents find what they need, not what's vaguely similar.

pipeline
FTSvectorRRFtruthbacklinkintenthandler
02

Typed knowledge graph

Every connection has semantic meaning — not just “related to.” People have roles. Companies have investments. Projects have owners. Decisions have preconditions. The graph knows the difference, and agents can traverse it with type-aware queries.

invested_inworks_atfoundedownsreports_todepends_onconflicts_withsupersedesmentions...any type
03

Graph intelligence

PageRank surfaces your most central entities. Louvain community detection finds natural clusters. Shortest path traces how any two nodes connect. Node similarity discovers structural twins. All powered by Neo4j GDS with automatic Postgres fallback.

pagerankMost central pages
GDS / degree fallback
communitiesNatural clusters
GDS Louvain
shortest_pathHow A connects to B
Pure Cypher
similar_pagesStructural twins
GDS / Jaccard fallback
04

Autonomous enrichment

The dream cycle runs nightly without any human input. It extracts wikilinks and timelines from page content, generates OpenAI embeddings for new chunks, reconnects orphan pages via semantic similarity, detects cross-page patterns, escalates important entities, and syncs everything to the Neo4j graph projection.

extractWikilinks + timeline from content
embedOpenAI text-embedding-3-small
orphansAuto-link via semantic similarity
patternsCross-page co-occurrence
tiersEntity importance escalation
graph syncPostgres → Neo4j projection
brainbase.belweave.ai/demo

Interactive 3D Knowledge Graph

Explore a real brain: search pages, traverse links, see the graph structure live

No signup required
See it work

Try it with
your own API key

The interactive demo shows a real knowledge graph — search pages, explore typed links, and run graph intelligence queries. Then grab an API key from the dashboard and connect your agents.

23
MCP tools
SDK + CLI
+ MCP
Polyglot
PG + Neo4j