ai
ai copied to clipboard
🤖 Provider API update - [email protected]
Breaking changes (removed endpoints, parameters, or changed behavior)
- POST /v1/documents/search
- ranking_metric: Request parameter is now an enum (RANKING_METRIC_UNSPECIFIED, RANKING_METRIC_COSINE_SIMILARITY, RANKING_METRIC_L2_DISTANCE) or null, replacing the previous free-form string. Requests with unsupported values will now be rejected.
New features (new endpoints, parameters, or options)
-
POST /v1/documents/search
- Added reranker parameter (optional):
- Type RECIPROCAL_RANK_FUSION with config: k (int32, optional), weights (float[] optional)
- Type RERANKER_MODEL with config: instructions (string, optional), model (string, optional)
- Added search_type parameter (optional enum): SEARCH_TYPE_SEMANTIC (default when omitted), SEARCH_TYPE_KEYWORD, SEARCH_TYPE_HYBRID
- Added search_multiplier parameter (optional int32): valid range [1, 100]; applies to HYBRID search only (defaults to 1)
- Added reranker parameter (optional):
-
POST /v1/responses
- Extended tool call output types supported in response items:
- web_search_call: includes action details (search with query and sources; open_page with source; find with pattern and source)
- file_search_call: includes results with file_id, filename, score, text
- code_interpreter_call: outputs include logs or image URL
- mcp_call: includes name, server_label, arguments, status, error
- custom tool call: includes call_id, id, name, input, status
- Extended tool call output types supported in response items:
-
GET /v1/responses/{response_id}
- Same extended tool call output types as POST /v1/responses:
- web_search_call, file_search_call, code_interpreter_call, mcp_call, custom tool call
- Same extended tool call output types as POST /v1/responses:
https://github.com/gr2m/ai-provider-monitor/releases/tag/xai%4011.0.0