OpenEvalDocumentation
GitHub
REFERENCE

CLI & file reference

The author-facing controls, in one place.

On this page

Eval file contract

FileRequiredUsed for
benchmark.tsYes · projectModels, judge, repetitions, concurrency
evals/<id>/prompt.mdYes · evalVerbatim candidate task
evals/<id>/judge.mdYes · evalDeclared metrics and grading criteria
evals/<id>/eval.tsOptionalWorkspace preparation and early stopping
evals/<id>/workspace/OptionalCandidate-visible starting files

Commands

openeval …Action
imageBuild the isolated candidate runtime
planPreview missing / changed work and cost estimates
runExecute the current plan
viewServe the bundled results viewer
retry <run> <eval-run>Retry a candidate execution
rejudge <run> <eval-run>Judge existing evidence again
snapshot <run> <name>Freeze the active scores into a named snapshot
add-models <run> --model <ref>Extend an idle aggregate with more models
merge-runs <target> <source>Merge results into one aggregate

Flags

FlagApplies toPurpose
--benchmark <dir>All benchmark commandsBenchmark directory; default current directory
--run <dir>run / planUse a specific result
--newrunStart a separate result
--model <ref>run / plan / add-modelsAdd a model
--only-model <ref>run / planLimit execution to a model; repeatable
--only-eval <id>run / planLimit execution to an eval; repeatable
--only-repetition <n>run / planLimit execution to a repetition; repeatable
--max-cost <usd>run / planScheduling budget for this invocation
--final-onlyrunDisable monitoring for this invocation
--port <n>viewViewer port; default 4173

Benchmark settings

SettingDefaultMeaning
modelsRequiredUnique provider/model references; optional #variant
judge.modelRequiredModel that grades the recording
repetitions3Candidate executions per eval / model
concurrency10Shared worker budget
candidate.timeoutMs45 minutesCandidate limit; cannot exceed 45 minutes
judge.timeoutMs10 minutesJudge execution timeout
candidate.websearch / judge.websearchexaSearch provider, or false
container.enginedockerdocker or podman
container.cpus / memoryMiB2 / 4096Per-container resources
Find in documentation