Graphs
A graph node can represent a person, an email address or another entity, so
nodes counts more than input rows. To add rows, connect your workflow’s output to a Lattice block that targets the graph. Build a graph from workflow output
Match an incoming row
POST /lattice/graphs/{graph_id}/match finds earlier records that resemble the supplied row. Use the original workflow’s column names in fields. Supply the values you have, including full text for columns such as a bio. Lattice ignores unknown or unsearchable columns. Matching is read-only.
Set SIXTYFOUR_API_KEY to a key from the graph’s team and GRAPH_ID to its ID. This example assumes the graph was built from rows with an email column.
A
same_person match means both records point to the same person in the graph or multiple strong signals agree. A possibly_same match has one strong signal, such as a rare identifier or copied text. A related match can reflect a shared employer or document template. Read the evidence before using a match to make a decision.
For a retry of the same match, reuse the original body and MATCH_KEY. The API can replay the original result without charging again during the 24-hour idempotency window. Generate a fresh key for a new match.
Look up a known record
Use the ID from your source row. URL-encode it as a single path segment.
The lookup includes up to 100 attached nodes and 20 linked records. Each list has its own
next_cursor. Pass it as cursor to the corresponding endpoint until it returns null. See Handling Errors for lookup errors and Credit Costs for pricing.
Labels
Label definitions belong to the graph. A label can take its value from a workflow column or accept writes through the API.
For example, define an API-managed label with
PUT /labels/review_status and the body {"values":["pending","reviewed"]}. To set it on a record, send this body to POST /records/labels:
applied and rejected. A batch can contain both. Rejections identify an unknown label, a column-managed label, a disallowed value or a missing record. If the record is still being ingested, submit the rejected item as a new request after it becomes available. For a label managed by the workflow, set from_column when defining it and supply one of its allowed values in that column.
Follow graph changes
GET /lattice/graphs/{graph_id}/events returns events oldest first. The first request starts at the graph’s first event. The feed supports record.added, record.linked and label.set. When a new record links to an earlier one, record_id identifies the earlier record and data.linked_record_id identifies the new one. Setting a label can also emit record.linked for an existing connection. In that case, data.linked_record_id identifies the record whose label changed.
type to select multiple event types. For linked records, linked_label accepts a filter such as review_status:reviewed. The limit controls how many record additions and label writes the feed examines (default 20, maximum 50). Each can also emit its 20 strongest links, so a response can contain more events than the requested limit. Use the linked-records endpoint to explore additional links.
Save next_cursor after processing the page and send it as since on the next poll with the same filters. If has_more is true, fetch the next page immediately. Otherwise, retain the cursor for your next poll. Use event_id to deduplicate deliveries if your consumer retries. Events can take several minutes to appear.