SearchStack

Remove a Contributor from a List.

Add a Coordinates field to a List, enabling geographical queries.

Remove the Coordinates field from a List. Breaking change: bumps the List version.

Start an asynchronous eval run: each query is searched against the pinned records and its hits scored by the judge. Returns a run_id to poll with EvalRunStatus plus an up-front estimate of the billed judge calls.

Get an eval set's score history, newest first, paged by an opaque continuation token.

Poll one eval run: lifecycle status, the environment it ran under (model/reranker/judge), the tally counts and rates, and the per-query rows.

Bootstrap an eval set from prior retrieve-then-judge runs: each source run replays its search as one eval query, and the pinned target and judge are taken from those runs (which must all share one target and one judge). Judge runs whose candidates were supplied explicitly are skipped.

Create an eval set against a pinned list/group version. The version is frozen to a concrete number; a pin fixes WHICH RECORDS are searched, not the results (records, embeddings, model and reranker are read live at run time).

Delete an eval set and cascade its runs and per-query rows.

List the eval sets configured for an Account. An eval set is a named relevance test: query cases pinned to a list/group version and scored by a judge, with a score history.

Get an eval set's configuration (pinned target, judge, context and query cases).

Extract text from PDF documents attached to a Search Result. Only available when the List's model declares a positive MaxDocumentPages limit.

Create a Facet. A Facet is a filterable field (e.g. brand, color, price, category).

Remove a Facet. Breaking change: bumps the List version.

Rename an existing Facet. Breaking change: bumps the List version.

List the distinct values of a Facet on a List version. Useful for building filter UIs before searching. Requires the search-result:read permission, like the search tools.

Get the Public API discovery catalog (IETF API Catalog draft) with links to OpenAPI, docs, manifest and examples.

Get the Public API self-describing manifest (auth, error codes, MCP integration guidance). Call once per session.

Get a curated library of worked request/response examples for the Search Stack Public API.

Add an IP address to a Group's access whitelist.

Add a List (at a version) to a Group.

Bump a Group's version: re-pins every member List to its latest version and advances current_version. A no-op when every member is already current.

Clone a Group into a new Group, copying its configuration and member Lists. By default members are re-pinned to their latest version.

Create a Group. A Group combines multiple Lists for cross-list searching.

Remove a Group.

Edit a Group's default search/suggest size and distance unit.

Get a Group's configuration, member Lists and IP whitelist.

Get the member Lists (and pinned versions) that a specific group_version searches.

List every searchable membership version of a Group. Pass one as group_version to search a frozen set of records.

Remove an IP address from a Group's access whitelist.

Remove a List from a Group.

Remove the embedding model from a Group, reverting it to keyword search.

Detach the reranker from a Group, reverting to score-based ordering.

Restore a historical membership version as the Group's new current_version (roll-forward, not rewind). Member Lists deleted since that version are skipped and returned in skipped_lists.

Free-text and/or vector search across a Group of Lists. Hits include 'list_name' so you can tell which list each hit came from.

Free-text and/or vector search across a Group of Lists. POST variant accepting a SearchOptions body.

Set the embedding model for a Group. Member Lists are re-embedded asynchronously.

Attach a bring-your-own reranker to a Group.

Typeahead suggestions across all Lists in a Group. Requires the 'list-group:read' permission.

Typeahead suggestions across all Lists in a Group. POST variant accepting a SuggestOptions body.

Transfer a Group and its member Lists to a different subscription.

Create a bring-your-own LLM judge: a post-search filter that runs your LLM over each candidate and returns a schema-validated verdict. The API key is verified against the provider on create.

Delete a Judge.

List the bring-your-own LLM judges configured for an Account (secret API keys omitted).

Start an asynchronous batch Judge run over a candidate set. Returns a run_id to poll with JudgeRunStatus; results are produced off the request path.

Poll a batch Judge run: status, tally (processed/passed/errored) and the per-candidate verdicts persisted so far.

Synchronously evaluate a Judge against a single sample candidate — the prompt/schema tuner. Does not persist a run.

Get a Judge's configuration (provider, model, instructions, output schema, pass field).

Create a List. A List is a collection of related Search Results.

Permanently delete a List.

Edit List defaults and settings.

Retrieve the details of a List.

Restore a historical schema version (facets, searchable fields, resources) as the List's new latest version (roll-forward, not rewind). Schema only — Search Result data is not versioned.

Permanently delete a media item from a media store.

Records a click-through on a Group search result for analytics, correlated to its query by query_id.

Records a click-through on a List search result for analytics, correlated to its query by query_id.

Find Search Results within a List that are vector-similar to a given seed Search Result. Use for 'related items' / 'more like this' flows.

Find Search Results within a List that are vector-similar to a given seed Search Result. POST variant accepting a RelatedOptions body.

Create a Resource. A Resource is a non-searchable, non-filterable field (often a URL).

Remove a Resource. Breaking change: bumps the List version.

Rename an existing Resource. Breaking change: bumps the List version.

Image-similarity search of a List using an absolute, publicly reachable image URL.

Image-similarity search of a List using a base64-encoded image. Use when the image cannot be fetched by URL.

Image-similarity search of a List using an absolute image URL. POST variant accepting a SearchByImageRequest body.

Image-similarity search of a Group using an absolute, publicly reachable image URL. Hits include 'list_name'.

Image-similarity search of a Group using a base64-encoded image. Hits include 'list_name'.

Image-similarity search of a Group using a base64-encoded image. POST variant accepting a SearchByImageBase64Request body.

Image-similarity search of a Group using an absolute image URL. POST variant accepting a SearchByImageRequest body.

Free-text and/or vector search of a List version. Returns a SearchResponse JSON document with a 'results' array and 'count'.

Free-text and/or vector search of a List version. POST variant accepting a SearchOptions body. Use when filter/radius strings are long enough to risk URL limits.

Create a single Search Result.

Permanently delete a Search Result.

Queue a background job to permanently delete every Search Result in a List that matches the filter. A filter is required (an empty filter would match the whole List). Returns 202 Accepted once queued.

Edit the properties of an existing Search Result.

Directly update the text and image vectors of an existing Search Result. The List must not be attached to a model.

Queue a background job to soft delete (set active to false on) every Search Result in a List that matches the filter. A filter is required (an empty filter would match the whole List). Returns 202 Accepted once queued.

Create a Searchable field.

Remove a Searchable field. Breaking change: bumps the List version.

Rename an existing Searchable field. Breaking change: bumps the List version.

Typeahead suggestions within a List version. Returns an array of dictionaries (each dictionary is a suggested Search Result's fields).

Typeahead suggestions within a List version. POST variant accepting a SuggestOptions body. Use when filter/radius strings are long enough to risk URL limits.