VectorDatabase
> Auto-generated from the OpenAPI spec. Method names shown are the manifest > (canonical) names; per-language casing is applied by each SDK.
listRecordVectorContext
GET /record/vector/context
Lightweight semantic-context lookup designed for AI agents. — Returns a compact list of records most semantically relevant to the query. * Custom field values are emitted in short notation only (value, id, created, updated, * customFieldInputId, customFieldInputName) — no full DB representation. Designed for * AI agents that need lean context before acting; planned for future MCP-tool wrapping * so the response shape stays trivial to transfor
| Parameter | In | Required | Description |
|---|---|---|---|
space_id |
query | yes | Tenant scope. Custom module derived from space.custom_module_id. |
query |
query | yes | |
limit |
query | no | Page size — max records per API call (default 100). |
filter |
query | no | Filter expression. Mutually exclusive with segment_id. |
segment_id |
query | no | Saved segment whose filter to apply. When present, the response includes fromSegment with the segment name. |
Returns: any · Cost: 1 API call.
listRecordVectorOutliers
GET /record/vector/outliers
Find Record records semantically furthest from a text query, scoped to one space. — Inverse of /similar: returns records whose embedded text fields are most * DISTANT from the query (cosine distance, descending). Same filter/scope rules as * /similar. Note: HNSW indexes do not accelerate distance-DESC queries; outlier * ranking is closer to a full scan over the filtered candidate set.
| Parameter | In | Required | Description |
|---|---|---|---|
space_id |
query | yes | Tenant scope. Custom module derived from space.custom_module_id. |
query |
query | yes | Free-text query to embed; ranking returns records UNLIKE this. Max 4000 characters. |
limit |
query | no | Page size — max records per API call (default 100). |
filter |
query | no | Filter expression. Mutually exclusive with segment_id. |
segment_id |
query | no | ID of a saved segment whose filter to apply. Mutually exclusive with filter. |
max_distance |
query | no | Cosine-distance floor in [0, 2]. Records whose best embedding is CLOSER than this are dropped. |
Returns: any · Cost: 1 API call.
listRecordVectorSimilar
GET /record/vector/similar
Find Record records semantically similar to a text query, scoped to one space. — Returns the top-N records from a single space whose embedded text fields are * closest to the query (cosine similarity, descending). Pre-filtering via filter or * segment_id runs BEFORE vector ranking so the candidate base is the same as an * equivalent GET /record?filter=... listing.
| Parameter | In | Required | Description |
|---|---|---|---|
space_id |
query | yes | Tenant scope. Rows from other spaces are excluded. The acting user must have read access. The custom module the records belong to is derived from space.custom_module_id — the caller does not pass it. |
query |
query | yes | Free-text query to embed and rank against. Max 4000 characters to keep the URL within Apache/NGINX default 8 KB limits. |
limit |
query | no | Maximum records to return. Defaults to 20. No hard upper cap at the API layer. |
filter |
query | no | Filter expression in the same syntax as GET /record?filter=.... Resolves to a candidate-record set via FilterManager before vector ranking. Mutually exclusive with segment_id. |
segment_id |
query | no | ID of a saved segment whose filter to apply. Mutually exclusive with filter. Segment must belong to module custom-module-item. |
min_similarity |
query | no | Cosine similarity floor in [0, 1]. Records whose best embedding is below this similarity are dropped at SQL level (HAVING). |
Returns: any · Cost: 1 API call.
