Skip to main content
EigenPal lets you define AI workflows as files, test them against datasets, and run them from the dashboard, CLI, SDKs, or API. Most teams use it for document processing: parse, extract, classify, transform, and route data. Workflows can also call APIs, read websites, run scripts, and branch on logic. The core workflow is:
1

Build

Define a workflow as typed steps, or use a Git-backed agent for tasks that need code and per-input adaptation.
2

Evaluate

Pair it with a dataset and evaluators that encode the expected output.
3

Deploy

Start runs from the dashboard, CLI, SDKs, or API. Each run stores inputs, outputs, and traces. Experiments add evaluator scores.
4

Improve

Add failed real inputs to the dataset, fix the workflow, and compare the new experiment before shipping.
Workflows, agents, datasets, evaluators, and run artifacts are the main objects you work with in EigenPal. See Eval-first development for a worked example.

​
Start here

Why EigenPal

What it is for and when to use it.

How it works

The build, evaluate, deploy, and improve flow.

Choose your path

Pick CLI authoring, SDK integration, or raw REST API calls.

Your first workflow

Scaffold, push, and run a workflow from the CLI in minutes.

Eval-first development

One example from initial workflow to regression test.

​
Explore

Design principles

How EigenPal handles tests, files, versions, and execution.

Workflow vs agent

When to use a fixed workflow and when to use a Git-backed agent.

Evaluations

Datasets, evaluators, and experiments.

Reviews

Human verdicts, issue status, corrections, and monitoring.

Artifacts as files

How workflows, datasets, evaluators, and agents round-trip.

Workflow steps

Every supported step type, with its config and output schema.

SDKs

Trigger workflows and agents from TypeScript or Python.

REST API

Raw HTTP routes, request bodies, response objects, and schemas.

CLI reference

The full eigenpal command surface.