promptfoo/Test LLM outputs

Run promptfoo evaluations on your prompts. Works with pull requests, pushes, and manual triggers (workflow_dispatch).

View on GitHub

Trust Signals

Scorecard Score
not yet scored
Maintenance Recency
Stale
License
None
namedescriptionrequireddefault
github-tokenGitHub token used to list pull request files and post pull request commentsyes
promptsList of prompt files to watch (glob patterns supported, one per line)no
configPath to a Promptfoo config fileyes
cache-pathPath to the Promptfoo disk cache directoryno
openai-api-keyOpenAI API Key. Falls back to OPENAI_API_KEY environment variable if not provided.no
azure-api-keyAzure API key. Falls back to AZURE_OPENAI_API_KEY environment variable if not provided.no
anthropic-api-keyAnthropic API key. Falls back to ANTHROPIC_API_KEY environment variable if not provided.no
huggingface-api-keyHugging Face API key. Falls back to HF_API_TOKEN environment variable if not provided.no
aws-access-key-idAWS Access Key ID. Falls back to AWS_ACCESS_KEY_ID environment variable if not provided.no
aws-secret-access-keyAWS Secret Access Key. Falls back to AWS_SECRET_ACCESS_KEY environment variable if not provided.no
replicate-api-keyReplicate API key. Falls back to REPLICATE_API_KEY environment variable if not provided.no
palm-api-keyPalm API key. Falls back to PALM_API_KEY environment variable if not provided.no
vertex-api-keyGoogle Vertex API key. Falls back to VERTEX_API_KEY environment variable if not provided.no
cohere-api-keyCohere API key. Falls back to COHERE_API_KEY environment variable if not provided.no
mistral-api-keyMistral API key. Falls back to MISTRAL_API_KEY environment variable if not provided.no
groq-api-keyGroq API key. Falls back to GROQ_API_KEY environment variable if not provided.no
promptfoo-versionVersion of promptfoo to usenolatest
working-directoryThe working directory to run promptfoo inno.
no-sharePass --no-share and override config-level sharingnofalse
use-config-promptsUse prompts from config filenofalse
env-filesComma-separated list of .env files to load (e.g. ".env,.env.test.local")no""
fail-on-thresholdFail the action if the evaluation success rate is below this percentage (0-100)no""
debugReserved compatibility input; use GitHub Actions step debug logging to display debug messagesnofalse
max-concurrencyMaximum number of concurrent API calls (default: 4)no4
no-tableRun promptfoo with --no-table flag to keep output minimalnofalse
no-progress-barRun promptfoo with --no-progress-bar flag to keep output minimalnofalse
no-cacheRun promptfoo with --no-cache flag to avoid reading or writing results to the disk cachenofalse
disable-commentDisable posting comments to the PRnofalse
workflow-filesChanged-file list for workflow_dispatch mode (newline-separated). If not specified, falls back to workflow inputs or auto-detection.no""
workflow-baseSingle base ref to compare against in workflow_dispatch mode (e.g., main, HEAD~1). Revision ranges like main..HEAD are not supported. If not specified, falls back to workflow inputs or HEAD~1.no""
repeatNumber of times to run each test (must be >= 2). Useful for non-deterministic LLM evals where a single run is noisy.no""
repeat-min-passMinimum number of repeated runs each test must pass. Requires repeat to be set. Must be <= repeat. Example: with repeat=3 and repeat-min-pass=2, each test must pass at least 2 of its 3 runs. no""
force-runForce evaluation to run even if no files changednofalse
namedescription
cache-size-mbPromptfoo cache size in megabytes after the latest metrics check
cache-file-countNumber of files in the Promptfoo cache after the latest metrics check