Skip to main content
EigenPal is built for AI agents to use, not just humans clicking a dashboard. The design rule that follows from that: every object in the platform is a portable artifact. Workflows, datasets, evaluators, and agent source are all plain files you can export, edit, diff, version in Git, and import back. Nothing is locked inside a UI. That is what makes the platform agent-friendly. An agent can pull a workflow, read it, change one step, push a new version, run it against a dataset, and read the scores, entirely through files and the CLI or API.

​
The artifacts

ArtifactFormatExport / import
WorkflowYAML (workflow.yaml)eigenpal workflow pull / push
DatasetFolder archive of exampleseigenpal workflow dataset pull / push
EvaluatorsYAML (evaluators.yaml)eigenpal workflow evaluators pull / push
AgentGit-backed source packageeigenpal agents clone / commit / push
Run outputJSON + output fileseigenpal runs get / artifacts
Every one of these round-trips. What you pull is what you can push back. A single-example dataset export re-imports cleanly, so an agent can carry one example between environments without special-casing.

​
Why it matters

  • Agents can author the whole thing. A coding agent works in files. Because workflows, datasets, and evaluators are files, an agent can build and improve them end to end, the same way it edits source code.
  • Version control for free. Artifacts live in your repo. Diff a workflow change in a PR. Roll back a dataset. Track evaluator history.
  • No lock-in. Export everything, inspect it, move it between local, staging, and production with the same commands.

​
The CLI is the surface

The eigenpal CLI is the agent surface for all of this: authoring, running, and inspecting from a terminal. The same commands a human runs are the commands an agent runs. See the CLI reference for the full command tree, and the skill install guide to drop the CLI reference straight into your coding agent.