5a5597ebb2d6

latestOpenAPI 3.1.0raw.githubusercontent.com2026-08-126178842.7 MB
evals
evals

List eval runs

Live operational data — always revalidate; do not rely on a client-side cache. Follow next_cursor to page; cursors stay valid across concurrent writes. Note that filters derived from other tables (outcome, score_min/score_max) are re-evaluated per page, so a run whose scores land mid-walk can be missed — it can never be returned twice.

get/v1/eval-runs

Query parameters

session_idstring uuid nullable
evaluator_idstring uuid nullable
agent_idstring uuid nullable
status'pending' | 'running' | 'completed' | 'failed' | 'errored' nullable
scope'live' | 'simulation' | 'test' nullable
outcome'passed' | 'failed' | 'no_verdict' | 'errored' nullable
score_minnumber nullable

Min normalized score, matched against EvalScore.value [0,1]. numeric-format evaluators store their signal in raw_value (value is null), so they are not matched by score_min/score_max.

Min normalized score, matched against EvalScore.value [0,1]. numeric-format evaluators store their signal in raw_value (value is null), so they are not matched by score_min/score_max.

score_maxnumber nullable

Max normalized score [0,1]; see score_min re: numeric evaluators.

Max normalized score [0,1]; see score_min re: numeric evaluators.

severity'info' | 'low' | 'medium' | 'high' | 'critical' nullable
channelstring nullable
created_fromstring date-time nullable
created_tostring date-time nullable
is_criticalboolean nullable

Filter on the run's is_critical_snapshot. Omit for all runs; false returns non-critical runs only.

Filter on the run's is_critical_snapshot. Omit for all runs; false returns non-critical runs only.

qstring nullable

Case-insensitive substring match on the agent and evaluator names, plus the run / session / agent / evaluator ids (exact UUID, or an 8+ character id prefix). Runs whose agent_id is null (legacy rows) cannot be matched by agent name.

Case-insensitive substring match on the agent and evaluator names, plus the run / session / agent / evaluator ids (exact UUID, or an 8+ character id prefix). Runs whose agent_id is null (legacy rows) cannot be matched by agent name.

limitinteger
cursorstring nullable

Response

Successful Response

next_cursorstring nullable
has_moreboolean required