Compass is a context engine that builds a temporal knowledge graph of your organization — entities, relationships, and documents across every system — queryable by humans and AI agents alike. Search it. Traverse it. Reason over it.
- AI-Native -- Composable MCP tools let AI agents search, traverse, assess impact, and read documents over your organizational graph.
- Knowledge Graph -- Typed, directed, temporal relationships between entities including lineage, ownership, and custom edge types.
- Context Assembly -- Multi-hop bidirectional traversal builds a subgraph around any entity — the full picture in one call.
- Impact Analysis -- Blast radius analysis traces what depends on what and what breaks when something changes.
- Hybrid Search -- Keyword precision with semantic similarity using Postgres-native full-text search and pgvector embeddings.
- Documents -- Attach any document to entities — runbooks, decisions, postmortems, design docs — all indexed for semantic search.
- Open Type System -- Any entity type, any edge type, any properties. Hardware, software, operations, people — anything fits.
- Quickstart -- Get running in 5 minutes
- Guides -- Entities, edges, search, context, MCP, CLI, API
- Internals -- Architecture, search engine, storage
Install Compass on macOS, Windows, Linux, or via Docker.
brew install raystack/tap/compassDownload .deb or .rpm from releases:
sudo dpkg -i compass_*.debdocker pull raystack/compass:latestgit clone https://github.com/raystack/compass.git
cd compass && make# Start PostgreSQL
docker-compose up -d
# Initialize and run
compass config init
compass server migrate
compass server start
# Search the graph
compass entity search "orders" --mode hybrid
# Explore context
compass entity context urn:bigquery:orders --depth 2
# Analyze impact
compass entity impact urn:kafka:events --depth 3Connect AI agents to Compass via MCP. Add to .mcp.json:
{
"mcpServers": {
"compass": {
"type": "sse",
"url": "http://localhost:8080/mcp"
}
}
}Compass is Apache 2.0 licensed.