PromptBeat and AgentBeat have separate product command interfaces. The website Eval Preview implements the current AgentBeat HTTP evaluator command; the older GitHub adapter Release does not.

​
PromptBeat commands

CommandPurposeRemote-call boundary
promptbeat validate --config <path>Validate a PromptBeat projectDoes not prove remote connectivity
promptbeat generate --config <project> --output <cases.json>Generate a deterministic local case previewNo Target or external-backend red-team call
promptbeat compile promptfoo …Compile project/input to backend YAMLLocal compilation
promptbeat run --config <project> --output-dir <dir>Run the full external-backend pipelineCan call Generator, Target, and Judge providers
promptbeat run promptfoo …Execute an existing backend YAMLCalls configured providers
promptbeat pipeline run …Explicit full-chain pipeline entryCalls configured providers
promptbeat eval …Evaluate backend YAML with optional Provider overridesCalls configured providers
promptbeat report …Render JSON or HTML from a resultLocal artifact generation
promptbeat dataset build-version …Build a versioned dataset artifact setCan generate, evaluate, and call providers
promptbeat dataset build-cycle …Run bounded multi-round dataset constructionCan generate, evaluate, and call providers
promptbeat seed pull …Pull a collection from a file/URL RegistryReads the selected Registry source
promptbeat config inspect …Inspect effective project configurationLocal configuration read
promptbeat scenarios list …List available scenariosLocal listing
See PromptBeat quickstart for a complete example.

​
AgentBeat command

The website-built Eval Preview supports agentbeat eval-run. It is not the older v0.3-agentbeat-preview.1 adapter Release, which only supports the legacy interface. Download and verify the current preview from the AgentBeat quickstart.
agentbeat eval-run < eval-run-request.json > eval-run-response.json
The website Eval Preview exposes this HTTP evaluation contract through the product command. It:
  • reads one strict JSON value from stdin;
  • requires AIBEAT_AGENT_TARGET_REGISTRY;
  • verifies request Target Profile/Capability against the selected Deployment;
  • calls the registered HTTP Agent through business-http-json-v1;
  • requires AIBEAT_JUDGE_BASE_URL and AIBEAT_JUDGE_MODEL;
  • emits one eval-run-response-v1 JSON value to stdout.
It can call a real Agent and LLM Judge and may produce Target side effects or model cost.

​
Source and release invocation

If your development checkout actually contains core/go, PromptBeat source commands run from that directory. The currently checked public GitHub snapshot does not include it; new users should use the full package in the PromptBeat quickstart. 
go run ./cmd/promptbeat --version
go run ./cmd/promptbeat --help
Do not insert an extra -- before --version. For an installed AgentBeat distribution, inspect its version and top-level help; a binary named agentbeat alone does not prove eval-run compatibility: 
agentbeat --version
agentbeat --help

​
Discover exact flags

Command help is generated by the installed build: 
promptbeat --help
promptbeat run --help
promptbeat dataset build-version --help
agentbeat --help

PromptBeat quickstart

AgentBeat quickstart