Overview

Pi (slug pi) is a coding agent CLI, driven non-interactively through its --print mode. It reaches its model through OpenRouter, so a Pi run reports OpenRouter-style model IDs and is priced from OpenRouter’s listed rates. See the Pi site for the harness itself.

Model IDs

Pi runs models through OpenRouter — every session is launched with --provider openrouter — so its model IDs are OpenRouter provider-prefixed slugs. Examples (illustrative, not exhaustive):

• z-ai/glm-5.2
• moonshotai/kimi-k2.7-code
• qwen/qwen3.7-plus

Invocation

The harness probes and invokes the pi binary. It is installed into the run container at run time with:

npm install -g --ignore-scripts @earendil-works/pi-coding-agent && npm cache clean --force

A session is run in --print mode with JSON output. The Test Cabinet passes these flags:

Flag	Purpose
--mode json	Emit the line-delimited JSON event stream consumed for events and usage.
--print	Run non-interactively, printing output rather than entering an interactive session.
--provider openrouter	Reach the model through OpenRouter.
--model <id>	The model to run.

The prompt is passed as the final positional argument.

Authentication. Pi’s API key is sourced from OPENROUTER_API_KEY on the host and passed straight through to the run container under the same name.

Pricing. Pi uses the Passthrough pricing model: the model ID is already an OpenRouter slug, so it is used as-is for the comparable-cost lookup. Pi does not self-report a run cost, so the comparable cost is always OpenRouter-derived; see Metrics.

---

See Events for how Pi’s output is normalized and Metrics for how its usage is counted. For the harness layer these pages fit into, see Harnesses.