Use case
Identity Resolution for AI Agents
AI agents writing to CRMs need to know whether a record already exists — and if so, which one. Without identity resolution, agents create duplicates, update the wrong records, or overwrite good data. Anchord resolves records to AnchorIDs using deterministic, rules-based scoring — not LLM guessing.
Why native CRM dedup isn't enough
Salesforce and HubSpot each catch duplicates within their own system. But neither sees records in the other system, or in Stripe, or in your data warehouse. An agent that writes a Stripe customer into Salesforce has no way to know the same company already exists in HubSpot under a slightly different name.
Cross-system identity is the gap. Every system knows its own records. No single system knows all of them. Anchord sits above these systems and maintains a canonical identity layer that spans all connected sources.
How Anchord solves it
Ingest records from any source — Salesforce, HubSpot, Stripe, or your own systems via the API. Anchord matches each record against existing entities using rules-based scoring signals: domain, email, provider IDs, name + location, and multi-signal corroboration. A three-tier pipeline — exact match, fuzzy reconciliation, human review — ensures nothing is guessed.
resolved High-confidence match to an existing AnchorID. The agent uses the returned entity ID to update the correct record in the target system.
not_found No match exists. The agent can safely create a new record in the target system without risk of creating a duplicate.
needs_review Multiple plausible matches exist and signals conflict. The agent stops, and the case routes to a human in the Anchord console for resolution.
The workflow: three API calls
The complete flow between your agent and the external system it's writing to is three calls:
Ingest
Push the record to Anchord via POST /api/v1/ingest/batch.
Or let built-in connectors pull records automatically via OAuth sync and webhooks.
Resolve
Call POST /api/v1/resolve/company
or /resolve/person.
Get back resolved, not_found,
or needs_review.
Guard
Before the agent writes, call POST /api/v1/guard/write.
Anchord evaluates the proposed write and returns allowed
or blocked with specific block codes. The agent decides what to do.
When identity is ambiguous
When Anchord can't confidently resolve a record — multiple candidates score within a narrow
margin — it returns needs_review.
The agent stops. A human reviews the candidates in the Anchord console, confirms the correct
match (or creates a new entity), and the agent retries. After resolution, the same resolve
call returns resolved with a
stable AnchorID.
This is safer than guessing. Every wrong merge or wrong update creates downstream problems in billing, reporting, and customer communication. Stopping on ambiguity costs a few minutes of human review. Writing to the wrong record can cost much more.
Resolve identity before every write
Sign up, get an API key, and make your first resolve call in under five minutes.