CogniGraph

An evidence-linked multi-model database in Rust. Store documents, connect them with typed relationships, and query across graph, full-text and vector search—with links back to supporting source text.

One server binary, Native storage backed by redb, and no external database or model service required for the example below. Website · Engineering docs · License

FOR a IN accounts
  FILTER a.case == @case
  FOR h, e IN 1..1 OUTBOUND a._id registered_on
    LET source = DOCUMENT(e.source_id)
    RETURN { account: a.handle, handset: h.label, evidence: source.text }

Community stores the relationships and evidence references your application supplies. Enterprise adds evidence-bound construction through Semantic Neurons and signed governance workflows.

Coming from ArangoDB? CGQL uses familiar AQL-style syntax for document queries, joins and traversals. The migration guide maps supported behavior and compatibility limits.

Five-minute start

Clone the public source and build the Community Docker image:

git clone https://github.com/cognigraphdb/cognigraph.git
cd cognigraph
docker build -t cognigraph .
docker run -d --name cognigraph -p 127.0.0.1:3000:3000 -v cognigraph-data:/data \
  -e COGNIGRAPH_AUTH_ENABLED=false -e COGNIGRAPH_EMBEDDING_PROVIDER=none \
  -e COGNIGRAPH_CGQL_MUTATIONS_ENABLED=true cognigraph
curl --fail --retry 10 --retry-connrefused --retry-delay 1 http://127.0.0.1:3000/health/database

The demo runs without authentication and is exposed only on loopback. For a shared deployment, configure authentication. See running for Docker Compose and production configuration, or the Helm chart for Kubernetes.

Build and run directly with Rust

Requires stable Rust and a C/C++ toolchain. From the cloned repository, build the server and run in a scratch directory to isolate its database and .env:

cargo build --release -p cognigraph-server
COGNIGRAPH_BINARY="$PWD/target/release/cognigraph-server"
COGNIGRAPH_DEMO_DIR="$(mktemp -d)"
cd "$COGNIGRAPH_DEMO_DIR"
COGNIGRAPH_HOST=127.0.0.1 \
COGNIGRAPH_PORT=3000 \
COGNIGRAPH_BACKEND=native \
COGNIGRAPH_NATIVE_PATH="$COGNIGRAPH_DEMO_DIR/cognigraph.redb" \
COGNIGRAPH_AUTH_ENABLED=false \
COGNIGRAPH_EMBEDDING_PROVIDER=none \
COGNIGRAPH_CGQL_MUTATIONS_ENABLED=true \
"$COGNIGRAPH_BINARY"

Leave the server running and use another terminal for the requests below. Bun is needed only for the optional React console.

Follow a relationship to its evidence

This small digital-forensics example is entirely synthetic: no real case, account, device or person. Start with a fresh demo database. Create an account, a handset and its extraction report, then link the account to the handset with an explicit reference to that report:

API=http://127.0.0.1:3000/api
H='content-type: application/json'

curl --fail-with-body -sS "$API/documents" -H "$H" -d '{
  "collection":"accounts", "_key":"rook_77", "handle":"rook_77", "case":"CS-0412"
}'
curl --fail-with-body -sS "$API/documents" -H "$H" -d '{
  "collection":"handsets", "_key":"8813", "label":"Handset 8813"
}'
curl --fail-with-body -sS "$API/documents" -H "$H" -d '{
  "collection":"extractions", "_key":"handset-8813",
  "text":"Accounts rook_77 and marla.v are registered on this handset."
}'
curl --fail-with-body -sS "$API/graph/relationships" -H "$H" -d '{
  "from":"accounts/rook_77", "to":"handsets/8813", "relation_type":"registered_on",
  "collection":"registered_on", "metadata":{"source_id":"extractions/handset-8813"}
}'

curl --fail-with-body -sS "$API/search/query" -H "$H" -d '{
  "query":"FOR a IN accounts FILTER a.case == @case FOR h, e IN 1..1 OUTBOUND a._id registered_on LET source = DOCUMENT(e.source_id) RETURN { account: a.handle, handset: h.label, evidence: source.text }",
  "bind_vars":{"case":"CS-0412"}
}'

The query response contains:

{
  "results": [{
    "account": "rook_77",
    "handset": "Handset 8813",
    "evidence": "Accounts rook_77 and marla.v are registered on this handset."
  }],
  "count": 1
}

Here, source_id is application-defined edge metadata and DOCUMENT() reads the referenced report. The application supplies the relationship; storing its source reference does not validate the claim or create a signed attestation. Prefix the query with EXPLAIN to inspect its plan. Use /api/query for CGQL mutations when enabled; /api/search/query accepts read-only queries. Both routes enforce their access controls.

Capabilities and editions

Community includes document storage, typed edges, BM25 text search, vector and hybrid retrieval, CGQL, sandboxed Lua, authentication, an admin CLI and the console. Native stores primary database state in a redb file; text indexes and optional vector sidecars are rebuildable derivatives. Enterprise artifact storage has separate backup requirements.

Edition	Available today	Planned
Community	Native storage, CGQL, search, Lua, auth, CLI, console, Docker and Helm	Read replicas and manual standby
Enterprise	Community capabilities plus Semantic Neurons, signed governance and multi-tenancy	Automatic failover, sharding and multiple writers

Today the server is a single-writer deployment. Cargo and Docker default to Community; build editions explains the Enterprise feature flag. ArangoDB is retained as a reference/conformance backend.

Community Components are free to run in production under the FSL-1.1-Apache-2.0, with conversion to Apache 2.0 two years after each release. Enterprise production use requires a commercial license. LICENSING.md defines the component split and use restrictions.

Rust workspace

Crate	Responsibility
cognigraph-core	Shared types and GraphBackend contract
cognigraph-query	CGQL grammar, validation, planning and execution
cognigraph-native	Native persistence, indexes, traversal and vector search
cognigraph-server	HTTP service and runtime orchestration
cognigraph-cli	Administration and offline repair
cognigraph-construct, cognigraph-governance, cognigraph-artifacts	Enterprise construction, authority and artifact custody

The architecture guide covers the other libraries. Start with CONTRIBUTING.md for development, local checks and the contributor agreement. AGENTS.md owns repository workflow rules.

Documentation

• Engineering index and implementation plan
• CGQL specification, AQL migration and HTTP API (also served at /openapi.yaml)
• Configuration, operations and DataOps
• HTTP and Lua examples, research evidence, issues and decisions
• Changelog and Enterprise license

Copyright 2026 Gedank Rayze, LDA.