What is new in Relata DB 3.0

Relata DB 3.0
Relata DB 3.0 is the current release. Pages that still describe the 2.x line carry a banner, and the 2.x documentation is kept in the 2.5.8 snapshot. Every claim here is limited by the recorded evidence listed in 3.0 evidence status.

Relata DB 3.0 is the open source edition (AGPL-3.0-only): a governed, bi-temporal knowledge database on one machine, with no clustering and none of the enterprise reliability and assurance layer. Clustering, high availability, several tenants, managed users and tamper-evident audit belong to Relata Enterprise DB, the next product line. See Editions.

What changes for you

AreaIn 3.0Read more
HTTP APIEvery public data-plane route moves under /v1/. Hard break, no aliases.Route restructure
Protocol doorsPostgreSQL wire, Arrow Flight, S3-compatible listener, Redis RESP, MongoDB wire, ClickHouse (HTTP and native), Neo4j HTTP, Bolt and the /sparql endpoint are retired. Stored data is kept.Migrating retired doors
Supported interfacesHTTP/JSON with an OpenAPI 3.1 document, read-only GraphQL (a translated subset), MCP, A2A, and search. SDKs: Python, TypeScript, Go.Interfaces
RetrievalOptional version-pinned citation records for query, search, GraphQL and MCP results, with replay.Retrieval evidence
InvestigationsAn end-to-end path from ingest to an archived, later retrievable case record.Investigation and archive workflow
MigrationA rehearsed (not qualified) SIEM/SOC cutover procedure and a supported-path matrix.SIEM/SOC migration
OperationsOffline backup, verify and restore; WAL generation rollover; capacity planning.Backup and restore, Capacity and WAL retention
GovernanceCell-level ACL, masking, purposes and a minimal local audit log in the open edition.Governance and audit

Editions

Relata DB (3.0)Relata Enterprise DB (next)Relata Cloud
SourcePublic, AGPL-3.0-onlyPrivate, signed binariesPrivate
Engine (bi-temporal storage, SQL, graph, search, vector, ingest)YesYesYes
HTTP/JSON, MCP, read-only GraphQL, Python/TypeScript/Go SDKsYesYesYes
Cell-level ACL, masking, classificationsYesYesYes
TenantsOneManyMany
Clustering, high availability, replicationNoYesYes
AuditMinimal local logTamper-evident, signed, exportableSame
Backup and restoreYesPlus point-in-time and disaster recoveryManaged

Route families that belong to the Enterprise edition (tenant administration, managed users, platform administration, tamper-evident audit proof and export, cluster operations) are not compiled into the open build, are absent from its OpenAPI contract and answer an enterprise-*-required problem response if called.

Current code versus target model
In the 3.0 source tree the clustering code is still in the repository, but the default build is single-node and omits the Enterprise route families. The server and cluster values of RELATA_PROFILE are being retired: the open build accepts only free. Do not read the table above as a claim that the repository split is complete.

Supported interfaces

  • HTTP/JSON under /v1/, described by the OpenAPI 3.1 document at GET /openapi.json. Probes (/health, /health/live, /health/ready, /version) and discovery stay unversioned.
  • GraphQL: POST /v1/graphql, read-only. A hand-written translator for a documented subset (field selection, limit, where, variables, restricted introspection), not a GraphQL engine.
  • MCP: POST /mcp (JSON-RPC 2.0, buffered JSON mode) and the /v1/mcp/* tool routes. Discover tools from the running server with GET /v1/mcp/tools.
  • A2A: /v1/a2a/tasks and /v1/a2a/checkpoints/*.
  • Search: POST /v1/search, POST /v1/search/multi, saved searches under /v1/search/saved.
  • Cypher subset: sent to POST /v1/query with the x-query-dialect: cypher header and translated to SQL. The Neo4j HTTP and Bolt listeners are retired.
  • Python, TypeScript and Go SDKs are the only supported consumer SDKs. Other languages call HTTP or generate a client from the OpenAPI document; a generated client is not a supported SDK. No Rust consumer SDK is supported.

Authentication is required: data-plane requests carry a token minted from the registry, not the bootstrap secret. Writes go through the ingest routes, never through POST /v1/query, which is read-only.

Data in: OpenETL

Relata DB stores, indexes and retrieves data. It does not fetch or transform it. Connectors (Kafka, TAXII, MISP), Sigma conversion, document preparation, model-backed extraction and media processing belong to OpenETL, a separate tool that loads prepared rows through bulk ingest. Relata may call one configured embedding endpoint for queries only. Several of these were removed from the tree in the 3.0 cycle; media processing is still partly present and is moving out.

What is not established yet

Read 3.0 evidence status before planning a go-live. In short: no workload is recorded as qualified, restart after very large ingests has failed on Linux in recorded runs, an in-place 2.7.1 to 3.0 data upgrade is not tested, and the SIEM/SOC migration has been rehearsed only on a synthetic fixture.