Skip to content

Factories > Harnesses

Configuring the Pi harness

Run Pi as a Warp Factories agent with a provider API key or authentication from your self-hosted worker.

Open in ChatGPT ↗
Ask ChatGPT about this page
Open in Claude ↗
Ask Claude about this page
Copied!

Pi is a coding agent that supports multiple model providers. Use it as a factory agent’s harness for tasks such as implementing changes through your team’s OpenAI account.

Choose a model from Pi’s catalog in the agent’s “Model” dropdown. Pi uses provider/model identifiers; available models depend on the selected credential or configured provider team defaults.

On Warp-hosted runs, Pi uses the selected model provider’s configured team default unless you select a different team-owned provider secret. Pi accepts compatible Anthropic and OpenAI API key secrets, as well as LLM provider API key secrets for other providers in its catalog. A personal key stored on your device is not a factory credential.

Your provider bills inference. Warp’s platform and hosted-compute charges still apply. Selecting Pi does not make every local Pi extension or feature available in a factory run.

You need editor or admin access to a factory, an existing factory agent, and a funded API account with a provider listed in Pi’s catalog. Follow harness plan requirements for access.

For a Warp-managed factory:

  1. Open the Warp Factories web app. Expand your factory in the sidebar, click Agents, then open the agent’s Settings tab.
  2. In the “Harness” dropdown, choose Pi.
  3. If you aren’t using a team default, in the “Auth” dropdown, choose a compatible team-owned provider secret. To create one, click New LLM provider API key, complete the provider, name, and API key fields, then click Create secret.
  4. In the “Model” dropdown, choose a model from the provider’s catalog. If you use a team default, “Auth” then displays Team default with the provider’s secret name.
  5. Click Save.

The saved settings show Pi, the selected model, and its authentication. Start a new task for this agent and inspect its run in the factory dashboard to confirm that the intended harness and model were used and that the provider accepted the credential.

For a GitHub-backed factory, edit the agent’s definition files instead. Set harness.type to pi, use a provider/model identifier in harness.model, and select the credential with harness.auth.source: managedSecret and harness.auth.secretName. See agentDefaults.harness for factory defaults and per-agent overrides.

Using authentication from a self-hosted worker

Section titled “Using authentication from a self-hosted worker”

If your self-hosted worker already has Pi’s provider authentication configured, use it without storing a provider secret in Warp.

  1. In the agent’s Settings tab, choose your self-hosted host in the “Host” dropdown.
  2. In the “Auth” dropdown, choose From worker environment.
  3. In the “Model” field, enter the worker’s model identifier in provider/model form, then click Save.

Start a new task for the agent to confirm the worker can authenticate to that provider. This option is available only with a self-hosted host; see self-hosting for worker setup.

If no models appear, confirm that the selected secret belongs to the team and matches a provider in Pi’s catalog, or that the intended provider has a configured team default. Changing to a different provider’s credential clears an incompatible model selection; choose the model again before saving.

If a run fails to authenticate, check the provider account’s funding and key permissions. For From worker environment, check authentication on the selected worker rather than creating a personal device key.