Trace Query

Command contract

The command sends POST /api/observability/traces/query. It uses the same target and credentials as other observability commands and targets https://observability.mastra.ai by default.

The inline JSON input is required. Before recommending the command, confirm that the installed CLI exposes it:

After confirming availability, inspect the target server’s current contract:

Use --schema to confirm that the target supports the route and to inspect the current request and response shape and structural constraints. Predicate paths are generic strings in the schema, so it does not provide the context-specific field/operator matrix. If you need more information, read the reference docs online (opens in a new tab). Do not infer unsupported predicates from storage columns or older documentation. The server remains the ultimate validation authority.

Query workflow

1. Run trace query --help and verify that the installed CLI exposes the command.
2. Run trace query --schema against the intended target to inspect the route contract and structural constraints.
3. Build the smallest query that answers the question. timeRange is required; from is inclusive and to is exclusive.
4. Keep page.limit small while exploring and project lightweight results with jq.
5. If data.page.next is non-null, repeat the identical query with that value in page.after.
6. Use trace get or trace span to inspect evidence for selected trace IDs.

Query a time range:

Find traces containing a failed tool call. Conditions inside one spans.some clause must match the same span:

Find traces with a low score from one scorer. Conditions inside one scores.some clause must match the same score record:

Pagination

The query response remains nested under data so the pagination cursor is preserved:

Pass a non-null cursor back without decoding or modifying it:

A cursor is bound to the accepted query and ordering. Do not change the time range, predicates, or ordering between pages. A mismatched cursor returns 409; a malformed cursor returns 400.

Predicate semantics

• Compose predicates with and, or, and not.
• Use some when one related record must satisfy the complete nested predicate.
• Use none when no related record may satisfy the complete nested predicate. Traces with no related records also match none.
• Related span, score, and feedback clauses correlate records by trace. Conditions in separate related clauses do not imply that they refer to one shared related record.
• String comparisons are exact and case-sensitive unless the canonical documentation states otherwise.
• The root timeRange filters the selected root span’s start time; it does not independently constrain related-record timestamps.
• Query results are lightweight and do not embed matching spans, scores, or feedback. Fetch trace or span details after selecting candidates.

Choosing the trace command

• trace list: browse recent traces with simple list filters.
• trace query: select completed traces using complex, explicit predicates.
• trace get <traceId>: inspect one selected trace.
• trace span <traceId> <spanId>: fetch one span in full.
• Trace Intelligence: discover aggregate recurring themes before drilling into individual traces.

Advanced queries only return completed traces. The configured observability store must support the query API. If the server rejects a query that follows the current structural schema and canonical documentation as unsupported, report that limitation rather than falling back to an inaccurate client-side approximation.