One agent run, 2.4 seconds, every promise kept.
The terminal on our landing page isn't decoration — it's the actual shape of a run. Here is claims.pre_inspection unpacked, line by line.
AGENT RUN · CLAIMS.PRE_INSPECTIONThe run, line by line
brain.run("claims.pre_inspection")
A capability is invoked — not a prompt, a named, versioned workflow with declared inputs, policies and gates.
connected — ERP · CRM · claims-db — 3 sources, RBAC applied
Stage 01: the run opens scoped connections to exactly three sources, with the caller's permissions applied before a single record moves.
structured — 9,412 records → knowledge graph
Stage 02: raw records resolve into entities — claimant, policy, vehicle, history — with lineage retained for every fact.
reasoned — policy v2.4 applied, rationale attached
Stage 03: the eligibility policy (version-pinned) runs against the graph. The decision carries its full reasoning.
decision: APPROVE · exception routed to a named owner
Stage 04: the in-policy approval proceeds; the one flag outside policy routes to a named human with context, not to a queue.
done in 2.4s — outcome written back to your system of record
Stage 05: the decision lands in the claims core. The graph learns; the loop closes; the metric moves.
From demo to your function
Swap the nouns and this is any operational decision you run at volume: KYC verification, invoice matching, quote pricing, ticket triage, layout generation. The pipeline is constant; the ontology, policies and skills are yours. That's what we mean by service as software — and it's two weeks from a conversation to a scoped run on your own data.
Want this run on your data?
Book a working session — we'll pick the function and show you the first run on your own records.