Three failure modes that make traditional identity APIs unsafe to act on inside an agentic pipeline.
Most identity APIs return a match/no-match. An agent that receives a boolean cannot distinguish "disqualified by evidence" from "insufficient evidence to qualify." Both look the same. Only one should block — and the agent has no way to know which is which.
When an agent makes a compliance decision based on identity data, the chain of reasoning should trace to citable evidence. If the identity API returns a black box, the agent's reasoning is also a black box. A regulator asking "why did the agent clear this transaction?" deserves a real answer.
An agent monitoring a customer for new sanctions exposure needs to know when something changes. Without standing queries, the agent polls — an expensive loop with false-negative windows between each check. A newly designated entity may remain undetected until the next poll fires.
Merge coherence, confidence, blocking evidence, and resolving attributes all arrive in the response envelope. An agent can branch on any of these fields — route on certainty, inspect blocking evidence before escalating, or surface the resolving attributes as justification for a positive decision.
Every claim the agent makes about an entity traces to a citable mention in the immutable evidence layer. The agent's chain of reasoning — "I cleared this transaction because matching attributes X, Y, and Z from sources A, B, and C" — is reproducible and auditable end to end.
Register a standing query at onboarding. When a customer's resolved entity changes state — a newly matching sanctions pattern, a previously uncertain cluster crossing a certainty threshold — the webhook fires. No polling, no false-negative windows, no re-running the workflow from scratch.
Kloom integrates with agent toolchains as a named tool: entity_coherence, evidence, alert_subscribe, curator_queue. Each tool call has a typed schema, a typed response, and a contract the agent can reason about. The API is designed to be consumed by an LLM-driven orchestrator, not just a human developer.
An initial resolve call returns merge_coherence of "likely" — consistent with a real match — but confidence of 0.61, below the threshold to act. A boolean identity API returns false. Kloom returns a structured response the agent can reason about: fetch more evidence, check what is missing, escalate with context.
Every step the agent takes — the initial resolution, the evidence fetch, the escalation — carries a citable mention_id. The agent’s chain of reasoning is auditable end to end against the append-only evidence record.
"Most identity APIs hand an agent a boolean. Kloom hands it a structured uncertainty surface — which is what an agent actually needs to act safely."
The cluster layer that generates the structured uncertainty surface. Coherence, certainty, blocking evidence, and resolving attributes are products of the probabilistic cluster model.
Read the pageThe immutable evidence layer that makes agent reasoning auditable. Every mention_id an agent follows traces to a citable, permanent mention record.
Read the pageThe primary use case for the agent surface: agentic onboarding and continuous monitoring workflows where standing queries trigger re-screening without manual intervention.
Read the pageThirty minutes with engineering. We run a live agentic screening workflow — resolve, fetch evidence, escalate — and walk the full uncertainty surface and audit trail at each step.