Recipes
Short, practical curl-based flows that match how agents actually integrate with Anchord. Each recipe includes prerequisites, step-by-step API calls, and expected success + failure responses.
1
Resolve Before Write
The most common agent pattern: ingest a record, resolve it to a stable AnchorID, guard-check, then write to your external system.
Steps
- Ping — verify API key
- Register source (if custom)
- Ingest record
- Resolve company or person
- Guard write
- Proceed (or escalate if blocked)
Covers: resolved, not_found, needs_review, LIMIT_EXCEEDED
2
Guard Before CRM Update
You already have an AnchorID entity from a previous resolve or sync. Before your agent touches the CRM, ask Anchord to evaluate the write.
Steps
- Ping — verify API key
- Get entity (optional — confirm it exists)
- Guard write
- Act on the decision (write or escalate)
Covers: allowed, blocked (NO_CANONICAL_LINK, LOW_CONFIDENCE), merged entity redirect
3
Needs-Review Escalation
Both resolve and guard can signal ambiguity. This recipe shows how to detect it, what to store, and how to route the user to the Review Queue.
Steps
- Resolve triggers needs_review
- Do not write
- Queue and link to Review Queue
- Retry after human resolution
Covers: needs_review, AMBIGUOUS_CANDIDATES, guard blocks, Review Queue actions
Common patterns
Every response includes
a
request_id
for log correlation and support.
needs_review means
Anchord found multiple plausible matches and can't auto-resolve. Don't
write — send the user to the
Review Queue.
Merged entities are
handled transparently — the API returns the canonical entity with a
redirect field.
Rate limits are
120 req/min per tenant. Plan quotas return
429 LIMIT_EXCEEDED
with usage details.
Scenarios
For workflow-level guidance on when and why to use each recipe, see the scenario pages: