Agent memoryby ZySec AI

Give your agents memory
that actually means something.

Long-term memory across sessions — governed, bi-temporal and traceable, and built on a real knowledge engine, not a scratchpad.

One binary · macOS · Linux · Docker · Kubernetes
0
Interfaces
0+
Canonical ID types
0
Agent interfaces
0
SDKs
Interfaces in 3.0
HTTP / OpenAPI 3.1Python SDKTypeScript SDKGo SDKMCP (Claude / Cursor)GraphQL (read-only)Arrow over HTTPOTLP ingestChange feedFiles APIHTTP / OpenAPI 3.1Python SDKTypeScript SDKGo SDKMCP (Claude / Cursor)GraphQL (read-only)Arrow over HTTPOTLP ingestChange feedFiles API
Standard interfaces
1
HTTP API for every client

If your app can call HTTP, you can use it.

Relata DB 3.0 exposes one HTTP/JSON API with an OpenAPI 3.1 document, Python, TypeScript and Go SDKs, MCP, read-only GraphQL and Arrow over HTTP. The Postgres, Redis, MongoDB, S3, Neo4j/Bolt, ClickHouse and Arrow Flight doors are retired; see the migration guide.

HTTP / OpenAPIPythonTypeScriptGoMCPGraphQL (read-only)Arrow over HTTP
Why Relata

AI forgets. Knowledge goes ungoverned.
Context gets lost across tools.

Relata is one database that fixes all three — memory with a backbone.

Memory that lasts

Agents remember across sessions — and can justify any decision with a traceable chain.

Answers you can trust

Provenance, time-travel and policy baked into every row and every query.

Open and well-specified

One HTTP/JSON API with an OpenAPI document, Python, TypeScript and Go SDKs, and MCP tools for agents.

Built for your AI journey

From one agent’s memory to a sovereign agent platform — same database.

Memory · governed tools · RAG · multi-agent · governance. One agentic database that meets you at rung one and scales with you all the way up.

The DIY stack6+ services · ETL glue · data drift
✗
Postgres— No time-travel. No graph. No identity resolution.
✗
Pinecone— Separate service. Separate bill. No governance.
✗
Neo4j— Another database to run. ETL breaks.
✗
Memory store— Ephemeral. No provenance. Forgets on restart.
✗
Audit log— Separate pipeline. Forgeable after the fact.
✗
Tool layer— Ungoverned calls. No ACL. No masking.
Replaced by
RelataDB
one binary · one query plane
temporalgovernedagent-nativeembeddingsbranchesaudited
Every capability the DIY stack provides — time-travel, provenance, identity, graph, search, embeddings, memory, audit, policy — in one engine. One engine to run instead of six services.
Use cases & industries

Built for high-stakes data.

Governed, temporal knowledge for security, finance, evidence and AI workloads, on one machine. Each use case below states what was exercised and what was not.

Exercised

Agent memory

Hover to expand →

Exercised

Governed RAG

Hover to expand →

Exercised

Video evidence

Hover to expand →

Pattern

Blockchain threat intel

Hover to expand →

Exercised

Financial crime

Hover to expand →

Exercised

Erasure and backup

Hover to expand →

Exercised

SIEM / SOC

Hover to expand →

Exercised

Clinical registry

Hover to expand →

How it works

Watch a memory form, then answer.

Data enters the knowledge graph, connections form, and when the agent is asked — it recalls the exact path and answers with provenance.

RelataDB · Agent Memory
AliceAcme Corp$2M txnRisk: HIGHMet Jul 15Flagged
What’s Alice’s risk profile?
HIGH risk — flagged for a $2M anomalous transaction linked to Acme Corp (Jul 15).Source: ingest #1842 · provenance verified
IngestConnectAskRecallAnswer
One database, many workloads

One engine, many workloads.

Hover to explore why one database is enough for memory, search, graph, governance and more.

Why RelataDB

Not just another memory layer.

mem0 and cognee are great memory SDKs. RelataDB is the knowledge engine underneath — with temporal history and governance they don’t offer.

Capability
RelataDB
Recommended
mem0
cognee
On-prem / sovereign
Partial
Bi-temporal history
—
—
HTTP/OpenAPI + 3 SDKs + MCP
Partial
Partial
Identity resolution
—
—
Policy in query
—
—
Local audit log
—
—
Knowledge graph
Partial
In-database embeddings
—
—
Schema ontology
—
Partial
Agent memory (MCP)
Open source
AGPL-3.0
OSS
OSS
Sovereign & enterprise

Built to run your most confidential data — on your premises.

Governance, provenance and policy from the ground up. Deploy on-prem, air-gapped, or in your own environment — your data stays under your control.

Governance

PURPOSE + cell-level masking on every read and write, every door.

Provenance & history

Bi-temporal rows with provenance — what the system believed, when, and why. Tamper-evident audit is Enterprise.

Portability

One binary on one machine: laptop, container, on-prem or air-gapped. Clustering is Enterprise.

Compliance

Erasure with receipts, legal hold, backup and offline restore.

Deploy anywhere

Run it your way.

One binary on one machine — laptop, container or server. No external database, no GPU. Clustering is Relata Enterprise DB.

DockerBinary · curl installDocker ComposeOn-prem / air-gapped
Pricing

Start free. Scale when you do.

Relata DB 3.0
Open source
AGPL-3.0-only · one machine · one tenant
  • Bi-temporal storage, SQL, graph, search, vector
  • HTTP/OpenAPI, MCP, read-only GraphQL, Python/TypeScript/Go SDKs
  • Cell-level access control and masking, local audit log
  • Backup and offline restore, erasure with receipts
Get started →
FAQ

Questions, answered.

How is RelataDB different from mem0 or cognee?

mem0 and cognee are memory SDKs. RelataDB is the knowledge engine underneath — a real database with bi-temporal history, provenance, identity resolution, and governance built in. Plus one HTTP API, three SDKs and MCP.

Is my data safe?

Sovereign by design. Deploy on-prem or air-gapped on your own machine. Policy, provenance, and cell-level masking apply to reads and writes. No data leaves your control.

Do I need to change my code coming from 2.x?

Yes, if you used the compatibility doors. Postgres, Redis, MongoDB, S3, Neo4j/Bolt, ClickHouse and Arrow Flight are retired in 3.0; use the HTTP API or an SDK. The migration guide maps each one.

Can I self-host?

Yes. It is open source and runs on one machine: Docker, bare metal, or macOS. Authentication is always required; the quickstart shows how to mint a token.

What does it cost?

Relata DB 3.0 is open source (AGPL-3.0-only). Relata Enterprise DB, with clustering, several tenants and tamper-evident audit, is priced separately: talk to us.

Does it support multi-agent swarms?

Yes. Agents get governed MCP tools and long-term memory (remember · recall · forget · justify), bi-temporal and access-controlled. Multiple agents share one knowledge plane on one node.