CLI overview - EigenPal Docs

Use the eigenpal CLI to author workflows, push datasets and evaluators, start runs, inspect traces, manage Git-backed agents, and configure organization email servers. Most resources are files, so the same commands work for local development, CI, and agent-assisted editing.

npm install -g @eigenpal/cli

Source is open at github.com/eigenpal/cli.

Authenticate to your organization

eigenpal auth login

This opens a browser to create an API key and stores it locally. One key maps to one tenant, so the key decides which organization you are working in. Work across several organizations by keeping a named profile per org and switching between them:

eigenpal auth list        # show configured profiles, active one marked
eigenpal auth use acme    # switch the active profile

Optional: install the skill

Install the EigenPal skill into a supported coding agent if you want it to work with EigenPal commands and file formats directly:

eigenpal skill install

The command opens an interactive picker for Claude Code, Cursor, Codex, and other tools. See Install the EigenPal skill for the full list and non-interactive options.

Read documentation from an agent or terminal

Every CLI release bundles the complete public documentation, the OpenAPI contract, and detailed agent references generated from the platform schemas. No browser, authentication, or network connection is required:

eigenpal docs search "evaluator path syntax"
eigenpal docs read reference/evaluators
eigenpal docs read steps/transform/script
eigenpal docs list api-reference --json

docs read writes the full source document to stdout, while docs list and docs search support --json for agent tooling. The bundle is release-matched, so its syntax and command reference describe the installed CLI version.

Output and polling flags

Pipeable data goes to stdout. Status, progress, hints, and errors go to stderr.
--json emits the command’s primary payload as JSON. --select <path> emits one nested value as JSON and is preferable to a separate jq process for stable paths.
--wait belongs on commands that start work; it blocks until terminal state. --watch follows work that already exists. Use --interval <seconds> and --max-wait <seconds> where exposed; a timeout exits with code 2.
--format chooses an export or generated-code encoding. It is not a synonym for --json.
runs get --expand requests server-side detail sections. --include is command-specific, so read that command’s help rather than assuming shared values.
When --out is omitted, export commands stream to stdout. Help identifies whether a particular --out expects a file or directory.

Common ways to work

Build and score a dataset. Create examples with expected output, push evaluators, and run an experiment from the CLI.

eigenpal workflow dataset push <workflow-id> --file ./dataset
eigenpal workflow evaluators push <workflow-id> --file ./evaluators.yaml
eigenpal workflow evaluator-type list
eigenpal workflow experiment run <workflow-id> --wait

See Build a dataset and Evaluate a workflow. Author and iterate on a workflow. Scaffold a project, push it, run it on a file, and read the trace to see what each step did. Edit workflow.yaml, push a new version, and compare experiments before deploying.

eigenpal init workflow parse-invoice
eigenpal workflow push
eigenpal run workflows.parse-invoice --input-file document=invoice.pdf
eigenpal runs trace <run-id>

See Your first workflow. Work with a Git-backed agent. Clone the organization’s agent source, edit it, then commit, push, and release.

eigenpal agents clone --out tmp/org-repos/acme-prod
eigenpal agents commit -m "teach invoice totals"
eigenpal agents push

See Agents.

Command reference

Every command group has its own page with all subcommands, flags, and arguments.

Command	Description
`eigenpal status`	One-shot dashboard: server, active tenant, user, key id, workflow count. Pair with —json for scripting.
`eigenpal init`	Scaffold a new workflow project. Without [name], scaffolds into the current directory using the cwd basename as the workflow name. With [name], creates ./<name>/ and uses that as the slug. The flat layout matches what run workflows.<slug> already discovers, no manual file moves.
`eigenpal auth`	Manage authentication. Credentials live in ~/.config/eigenpal/credentials.json as named profiles. Switch tenants with auth use <name> or set EIGENPAL_PROFILE=<name> for one shell.
`eigenpal models`	Inspect models and providers configured for the current tenant environment (text, vision, OCR).
`eigenpal email-servers`	Manage outbound email servers (ems_…): list, inspect, create, update, delete, and send a real test message.
`eigenpal workflow`	Manage workflows: push, pull, and evaluate.
`eigenpal agents`	Manage EigenPal agents: Git source, datasets, experiments, sessions, and releases.
`eigenpal run`	Start a workflow or agent run, e.g. workflows.extract-invoice.
`eigenpal rerun`	Create a new run from a previous run’s stored input snapshot.
`eigenpal runs`	Inspect, watch, and manage workflow, agent, and eval runs.
`eigenpal skill`	Install / uninstall / list the EigenPal agent skill in your project (Claude Code, Cursor, Codex, Gemini CLI, Antigravity, OpenCode, Pi, Windsurf, GitHub Copilot).
`eigenpal docs`	List, search, and read the complete release-matched EigenPal documentation offline.

​Authenticate to your organization

​Optional: install the skill

​Read documentation from an agent or terminal

​Output and polling flags

​Common ways to work

​Command reference