BrainAPI is a knowledge graph–powered AI memory layer that transforms unstructured data into structured knowledge, enabling intelligent search, recommendations, and contextual memory for AI agents and applications.
Production validation artifact contract
Run scripts/check_release_readiness.py DIR before creating an RC or stable
tag. DIR must contain the following files produced against the exact
candidate image:
smoke.json:lightandheavy, each withunexpected_5xx: 0and true flow flags for authentication, brain creation, plain/structured ingestion, task completion, context, search, Console, and MCP.latency.json: context/search p50, p95, and p99; context records zero online LLM retrieval loops; search recordsexcludes_embed_query: true.restore-light.jsonandrestore-heavy.json: matchingbefore/aftercounts for brains, nodes, edges, chunks, vectors, and observations plus ten deterministic retrieval checks withmatch: true.security.json:high: 0,critical: 0, thesha256:image digest, and the relative path of its SBOM.
The public ledger must also contain a representative LongMemEval result and a WANDS representative Search row. LoCoMo and BEAM rows retain their explicit sample scopes and are not promoted into broader claims by this gate.
Dev tags (vX.Y.Z-dev / vX.Y.Z-dev.N) still require smoke, restore, and
security on the exact image. They skip the public-ledger and latency SLO
claims so a working image can publish before those product scores exist.
Pass --tag so the publish workflow can apply that distinction.
scripts/production_smoke.py exercise produces profile-scoped smoke, latency,
and pre-restore state evidence; verify-restore produces the matching restore
report. scripts/assemble_release_artifacts.py combines successful light and
heavy workflow artifacts. The tag-only publish workflow performs this assembly
and runs the gate itself, in addition to requiring all quality and dedicated
heavy-runner checks on the tagged commit.