WormBase is an agent-orchestrated, ledger-substrated data platform. It installs into your chat in 60 seconds, grows your lake agentically from six trigger flows, compounds knowledge across eight lake-side loops, and answers every question with hash-chained receipts. Probabilistic edge. Deterministic core. Replays bit-exact from t = 0.
Installs into your Slack workspace, auto-discovers your team across
channels, and ingests files dropped in conversation. WhatsApp in
preview; Discord and Teams via the same ChannelAdapter
contract. Lurks by default; speaks only when a relevance gate fires.
Six source-building flows: drop‑and‑profile, credential‑in‑DM,
mentioned‑in‑conversation, dashboard form, KPI‑gap‑triggered,
lake‑discovery. Twelve plus connectors behind a single
Connector
Protocol. Bronze → silver → gold replays bit-exact.
Lineage, quality checks, schema-evolution impact, semantic-type fingerprinting, column classification, entity stitching, source-candidate triage, catalog drift. Cross-axis chains reinforce each other. Synthesis happens at ingest, not at retrieval.
Detect · classify · reinforceEvery answer carries the ledger entries that produced it. Hash-chained, replayable, auditable. Wire-replay feeds recorded infrastructure events back through the channel adapter for deterministic re-runs — same code path, different input source.
Trace · verify · replay
WormBase reduces to a single write primitive and a single read
primitive. The write primitive is PEVR:
propose → execute → verify → resolve.
Every persistent state change — a profile, a KPI, a person
discovery, an answer — flows through one of those four entry
shapes.
The ledger is append-only, company-scoped, hash-chained. It is the substrate, not a side-channel. Lake, memory, KPI tree, channel history, gate outcomes are materialized projections over the ledger. Replay it to timestamp T and you reproduce any state — same hashes, same bytes.
The conversational edge is LLM-driven and stochastic by design. The core writes are deterministic. The gate between the two is where trust is enforced. We do not trust the LLM — we trust the gate between the LLM and the substrate.
Governance follows the same shape: Person × Domain × Resource × Classification × Policy, all materialized from the ledger. No separate database, no portal, no workflow engine. Governance is code.
WormBase runs as a small set of services: the worm-core hub, a Postgres for ledger and projections, an inference endpoint, and a channel-adapter for each connected platform. The compose file wires them together; you bring the OAuth credentials and the first data source.
# clone the repo $ git clone https://github.com/wormbase/wormbase.git $ cd wormbase # bring up worm-core, ledger, inference, channel-adapter $ docker compose up # dashboard at http://localhost:3000 # connect a chat platform, then a data source
Today, contributors run packages directly from the workspace
(make test-all,
make qa).
See DEVELOPERS.md for the contributor path; the hosted compose
path lands when the OSS release ships.