— Proof

Every claim, with its evidence

This registry is the source the website copy is checked against. A claim without evidence is marked planned or not supported, never stated as fact.

Last reviewed 2026-10-28 against agentbrain-app main @ c0af5664

How to read the labels
  • Claim: Connectors to databases, files and object storage
    Status: Implemented
    Evidence: internal/connector/* (pg, mysql, bigquery, d1, mongodb, sqlserver query-only, sheets, drive, s3, r2, gcs, redis, http)
    Stated on: / · /features
  • Claim: Knowledge base with full-text and semantic search
    Status: Implemented
    Evidence: knowledge base: PostgreSQL FTS + vector rerank
    Stated on: / · /features
  • Claim: Knowledge graph with provenance and validity dates
    Status: Implemented
    Evidence: KG entities with valid_from / valid_until and extraction provenance
    Stated on: / · /features
  • Claim: Org and resource permissions with department scope and row filters
    Status: Implemented
    Evidence: resource_permissions: users, roles, groups, department subtree, row filters
    Stated on: / · /features · /security · /trust
  • Claim: Backend permission checks for MCP requests
    Status: Implemented
    Evidence: MCP server delegates checks to backend; agent_task / workflow_run via MCP not supported
    Stated on: /features · /trust
  • Claim: ETL and data workflows
    Status: Tested
    Evidence: ETL workflow engine, 107 automated tests
    Stated on: /features
  • Claim: Human approval checkpoints in workflows
    Status: Implemented
    Evidence: workflow approval checkpoints (approve / reject / cancel)
    Stated on: /features · /trust
  • Claim: Append-only audit log
    Status: Tested
    Evidence: audit table REVOKE + triggers, integration tests; no hash chain
    Stated on: / · /security · /trust
  • Claim: Configurable audit retention and legal hold
    Status: Implemented
    Evidence: retention 30–3650 days per org, legal hold
    Stated on: /security · /trust
  • Claim: AES-256-GCM encryption of stored credentials
    Status: Implemented
    Evidence: AES-256-GCM, single application key
    Stated on: /security · /trust
  • Claim: AI egress monitor, redact and block modes
    Status: Implemented
    Evidence: AI egress monitor / redact / block per org
    Stated on: /security · /trust
  • Claim: Managed cloud deployment
    Status: Production-supported
    Evidence: managed deployment on internal Kubernetes
    Stated on: /security · /trust · /pricing
  • Claim: Packaged deployment on your infrastructure
    Status: Planned
    Evidence: container image exists; no Helm chart or on-prem guide
    Stated on: /security · /trust · /pricing
  • Claim: SAML / OIDC / LDAP federation
    Status: Planned
    Evidence: login via builder-auth only
    Stated on: /security · /trust
  • Claim: SIEM export and OpenTelemetry tracing
    Status: Planned
    Evidence: Prometheus metrics only; no audit export or OTel
    Stated on: /security · /trust
  • Claim: Air-gapped operation
    Status: Not supported
    Evidence: builder-auth and default AI providers require network
    Stated on: /security · /trust
  • Claim: Customer-managed keys (BYOK / KMS)
    Status: Not supported
    Evidence: no per-tenant or customer-managed keys
    Stated on: /security · /trust
  • Claim: Compliance certifications or attestations
    Status: Not supported
    Evidence: no third-party assessment of any kind
    Stated on: /security · /trust
  • Claim: Slack, Teams, LINE or Zalo channels
    Status: Not supported
    Evidence: outbound Lark / Telegram notifications only
    Stated on: / · /features
  • Claim: Plan tiers enforced by license in product
    Status: Not supported
    Evidence: per-org feature flags set by our team; no license enforcement
    Stated on: /pricing
  • Claim: Policy-aware answer demo on the homepage
    Status: Illustrative
    Evidence: design illustration on synthetic data; no end-to-end fixture
    Stated on: /
  • Claim: Agents from Dewee reading AgentBrain context
    Status: Separate product
    Evidence: Dewee is a separate product; connects via MCP per deployment
    Stated on: / · /pricing
Proof — AgentBrain