tools/list.
To see exactly what your client discovered, ask it to list its tools or query the endpoint’s
tools/list method. Parameters map to a PolyAI REST endpoint — they’re sent as path segments, query parameters, or a request body — but your client handles that wiring from the schema; the tables below just list what you can set.Search & read conversations
Find conversations by their metrics, search across transcript text, and pull back the full turn-by-turn record.search-conversations
Search conversations by metric filters, with sorting and pagination.
Each conversation in the response carries a flat
metrics map and a custom_scores list. AI Scores appear in both: metrics holds each one’s 1–5 value under its CUSTOM_SCORE_<ID> key, and custom_scores holds the same value together with the breakdown.
search-transcripts
Full-text search over transcript turns across conversations, returning matches with surrounding turns for context.
get-conversation
Fetch every turn of a single conversation.
Metrics & analytics
Discover which metrics you can filter and aggregate on, then roll one up over time into buckets you can chart.get-metrics
List the filterable metrics for an account, optionally scoped to a project — built-in metrics plus any custom metrics the project defines. With a project scope, it also lists the project’s AI Scores as CUSTOM_SCORE_<ID> metrics, each with the score’s name as its display_name. Use this first to discover what you can filter and aggregate on.
aggregate-metrics
Aggregate a metric over a time window into buckets you can chart — with grouping, cohort filters, and post-aggregation thresholds.
This tool is experimental — its schema may change. Your client discovers the current name from
tools/list.Traces & function logs
Find a conversation’s spans, break latency and errors down per turn and per span, and read the log lines your GenAI functions write. These tools wrap the Traces API; see it for the span names, measures anderror_type values.
list-product-conversations
List the conversations that produced spans in a time window, newest first, with turn and error counts.
get-product-trace
Every span of one conversation, ordered by start time, with parent_span_id so the tree can be rebuilt. Function spans carry the span_id to pass to search-function-logs.
get-product-timeline
One entry per turn and span name with start, end, duration and error count — a per-turn latency picture without every span.
get-product-turn
The spans of one turn of a conversation.
search-product-spans
Find spans across conversations by name, status, duration and window, newest first. For counts and percentiles use aggregate-product-spans.
The three tools below are experimental — their schema may change. Your client discovers the current names from
tools/list.aggregate-product-spans
Percentiles, counts and error rates over spans, bucketed by time and grouped by span name, turn, function, error type and more.
search-function-logs
Page through the log lines your GenAI functions wrote, newest first. Returns each line’s event label and the span it ran in, never the raw line or its payload.
Each line carries
timestamp, level, function_name, conversation_id, event, error_type, trace_id, span_id and turn_index. The response echoes the from/to it used; pass them back when paging.
aggregate-function-logs
Count log lines and the conversations they belong to, bucketed by hour or day. Takes the same filters as search-function-logs.

