Governed · Bi-temporal · Ontology-native

The knowledge layer
AI agents can trust

Bi-temporal history, OWL 2 ontologies, identity resolution, and provenance built into every query. For AI developers, researchers, and knowledge engineers.

Get started →Explore the docs
Python · SQL · TypeScript — any client, any language
# Give your AI agent governed memory in 3 lines
from relata import Memory

m = Memory('http://localhost:9090', purpose='agent')
m.add('Alice's risk score: HIGH — flagged 2026-01-15')   # REMEMBER
hits = m.search('Alice risk flags', as_of='2026-01-01')  # RECALL AS OF
# Every memory: bi-temporal + governed by ACL + hash-chained provenance

──── or SQL ────

-- Resolve an entity, travel back in time, prove it
SELECT RESOLVE_IDENTITY(entity_id) AS canonical,
       event_type, confidence, prov_author
FROM AgentMemory
AS OF '2026-01-15T09:00:00Z'
WITH PROVENANCE
WHERE agent_id = 'researcher-7'
  AND confidence > 0.85
ORDER BY valid_from DESC;
How it works

From raw data to governed knowledge

Five stages. No manual wiring. Every fact temporal, governed, and queryable.

CSV / JSON
Postgres
Kafka / Events
S3 / Files
HTTP API
01
IngestAny protocol — Postgres wire, HTTP, Kafka, S3, gRPC, or direct SDK ingest
02
IdentifyRESOLVE_IDENTITY links records via 75+ canonical types: email, IBAN, IMEI, MMSI…
03
Build OntologyOWL 2 QL/EL graph assembles automatically; relationships inferred, SPARQL endpoint live
04
Record HistoryEvery mutation stamped with valid-time + system-time; bi-temporal AS OF queries enabled
05
Govern & ServeACL + cell masking + PURPOSE + hash-chained PROV-O; agents query via MCP, SQL, SPARQL
MCP RECALL
SQL AS OF
SPARQL
Arrow Flight
A2A Protocol
Ontological graph building live — entities link across sources automatically
memberOfparticipatedInownsoccurredAtholdsPersonOrganisationEventAssetLocation
Person
Organisation
Event
Asset
Location
Ontological relationship
What only RelataDB shipsOperators and verbs no other database has
AS OF 'timestamp'
Time-travel any tableBi-temporal history built into the row model. Valid-time and system-time on every record.
WITH PROVENANCE
Cryptographic audit trailHash-chained commit manifests. Who changed what, when, and why — unforgeable.
RESOLVE_IDENTITY()
Cross-source entity matching75+ canonical identifier types. Match the same person across email, IBAN, IMEI, MMSI.
IDENTITY_CLUSTER()
Probabilistic entity clusteringGroup records that likely refer to the same real-world entity across disparate sources.
REMEMBER / RECALL
MCP memory verbs for agentsNative MCP tools: remember, recall, reflect, consolidate, forget, associate, resolve.
PURPOSE 'id'
Governance in the query pathEvery query carries an audited purpose. Cedar-inspired ABAC + cell masking enforces it.
OWL 2 QL/EL
Ontological reasoning built inSchema-as-code with git-branched ontology. SPARQL endpoint and PROV-O graph included.
HYBRID_SCORE()
BM25 + vector + identity fusionCustom BM25, HNSW, and identity graph fused via reciprocal-rank scoring — one query.
Who builds with RelataDB
AI Developers

Drop-in knowledge backend for your AI stack

  • REMEMBER / RECALL / FORGET — native MCP memory verbs
  • Postgres wire, HTTP, gRPC, Arrow Flight — any client works
  • SDKs: Python, TypeScript, Go, Rust
  • A2A protocol + LangGraph adapter out of the box
  • Per-agent, per-tenant session isolation built in
Get started →
AI Researchers

Temporal reasoning and provenance on every fact

  • AS OF queries — valid-time and system-time dimensions
  • Hybrid retrieval: BM25 + HNSW vector + IDENTITY_CLUSTER
  • WITH PROVENANCE — hash-chained audit on every write
  • Governed agent sessions with confidence and recency scoring
  • Continuous anomaly detection jobs over stored data
Get started →
Knowledge Engineers

Ontological structure with full graph semantics

  • OWL 2 QL/EL reasoning engine — no plugin, no external dep
  • Live SPARQL endpoint over all stored types
  • RESOLVE_IDENTITY + IDENTITY_PATH operators
  • PROV-O provenance graph on every commit
  • Git-branched ontology with HEAD pointer and diff history
Get started →
Core capabilities
T±

Bi-Temporal History

valid_from/valid_to + system_from/system_to on every row. AS OF queries, no triggers, no shadow tables.

∩ID

Identity Resolution

75+ canonical types — email, IBAN, IMEI, MMSI, mobile money. RESOLVE_IDENTITY, IDENTITY_CLUSTER, IDENTITY_PATH.

ACL

Governance & Provenance

Cedar ABAC, cell masking, PURPOSE audit, hash-chained PROV-O manifests. Deny-wins, fail-closed.

MCP

Agent-Native Memory

MCP memory verbs + A2A protocol. Cognitive verbs map directly to governed temporal storage rows.

FTS

Hybrid Search

Custom BM25 + q-gram + HNSW vector + identity fusion. RRF scoring. No Tantivy. No external deps.

OWL

Ontology & SPARQL

OWL 2 QL/EL inference engine, live SPARQL endpoint, git-branched schema with HEAD pointer.

Connect from any tool or language
Postgres wiregRPCHTTP/RESTMCPArrow FlightSPARQLS3-compatRedis-compatClickHouse-compatNeo4j/BoltMongoDB-compatpgvector
Quick install
Running in 30 seconds

Single container, no external dependencies. Postgres wire on port 9090 — any psql client or SDK connects immediately.

# Run with Docker
$ docker run -p 9090:9090 ghcr.io/openworkbench-co/relatadb:latest

# Pick your SDK
$ pip install relata-sdk              # Python
$ npm install @relata/sdk             # TypeScript
$ go get github.com/openworkbench-co/relata-go

→ postgres://localhost:9090
→ http://localhost:9090/mcp    (MCP tools)
→ http://localhost:9090/sparql (SPARQL)
Read the quickstart →