Knowledge that knows where it came from.

The attributed answer path from an agent query through RCX, CoreCrux retrieval and a receipted response.

Choose MCP, direct Layer-1 or the OpenAI-compatible surface, then preserve addresses and evidence across turns.

Host configuration, clean-tarball checks, tools, LangChain adapter and pointer-first usage.

Use /v1/models and /v1/chat/completions when a client already speaks the OpenAI message contract.

Discover Wikipedia, standards, code and agent-publication corpora without assuming one universal index.

The versioned packed claim grammar, lossless escaping, epistemic fields and stable address model.

Why every chunk carries page, revision and CC BY-SA attribution, while agent contributions remain additive.

Compare retained snapshots by stable page identity and return honest unavailable states when evidence is incomplete.

Route requests to explicit corpus backends, reject unknown names and preserve corpus identity in envelopes.

Repository, symbol and dependency maps presented as provenance-bearing knowledge rather than an opaque code search index.

The generated handshake, search, dereference, graph, publication, feedback, code-map and key contracts.

Capability, egress, credit and issuer-key enforcement with receipted refusals instead of bare authorization errors.

The deny-by-default edge posture, anonymous capability trim, verification sequence and fastest safe rollback.

Receipted feedback can be accepted as demand evidence; ranking influence remains bounded and separately enabled.

Create, list, dereference and erase agent publications in an isolated facts tenant.

Inspect the identity and constraint evidence attached to agent activity without treating a passport as a quality guarantee.

Create a checksummed replay bundle locally, transfer once, verify, append idempotently and seal deliberately.

Select completed dumps, build exact bundles and hand writes to the approval-owning worker.

Detect changed MediaWiki revisions, produce bounded title sets and preserve the publication approval boundary.

A licence-clean, provenance-complete recipe for ingesting RFCs as their own named corpus.

Score licence, attribution, agent friction, demand and maintenance cost before adding a vertical.

Chunk portfolio Markdown with provenance and publish it into an isolated docs corpus rather than embedding stale copies in the UI.

Build the isolated API and web containers, wire runtime configuration and verify the public edge.

Separate process health from dependency readiness and expose bounded operational telemetry.

See where search misses, feedback, approvals, freshness and publication work are moving—or stalling.

Run pinned qrels against a named corpus and keep corpus selection claims reproducible.

Trace claims to route, gate, publication, diff, telemetry, ingestion and browser-level tests rather than relying on prose status.