Pi

Pi is a minimal coding-agent harness that you adapt to your own workflow. A community extension, pi-doubleword-provider, adds Doubleword as a model provider and lets you flip each request between the realtime tier and the cheaper async (flex) tier. This guide wires it up with GLM-5.2, a strong open weights coding model.

1. Install Pi and the Doubleword provider

If you don't already have Pi, install it with the script or via npm:

# macOS / Linux (install script)
curl -fsSL https://pi.dev/install.sh | sh

# npm (any OS)
npm install -g --ignore-scripts @earendil-works/pi-coding-agent

Then add the Doubleword provider extension:

pi install npm:@snehalyelmati/pi-doubleword-provider

n.b. This is a community extension and its page has the latest options.

2. Generate a Doubleword API key

1. Sign in to the Doubleword Console.
2. Go to API Keys and select Create API Key.
3. Copy the key immediately, since it's only shown once, then set it in your shell:

export DOUBLEWORD_API_KEY="your-doubleword-key"

The extension reads DOUBLEWORD_API_KEY from your environment, so there's nothing else to configure to authenticate.

3. Run Pi on Doubleword

You can point Pi at a Doubleword model by prefixing the model id with doubleword/ to smoketest the connection:

pi --model doubleword/zai-org/GLM-5.2-FP8

To use Pi interactively, type pi in your terminal.

Once Pi is running you can switch models mid-session with /model or Ctrl+L, so you can keep GLM-5.2 for heavy coding and drop to a smaller model for quick edits. To check the integration, ask it to read part of your project, for example "summarize the architecture of my src directory".

4. Switch between realtime and async

Press Alt+T or run /doubleword-mode to toggle between the two tiers, or set the mode up front:

export DOUBLEWORD_EXECUTION_MODE=async   # realtime (default) or async

Realtime answers straight away and suits interactive back-and-forth. Async runs on the flex tier at a lower rate, which fits long background edits and repository-wide passes where a minute of latency is a fair trade for the saving. You can also press Shift+Tab to change reasoning depth or turn reasoning off.

Optional settings

The extension reads a few more environment variables if you want to tune it:

For the full list of supported models and pricing, see the Doubleword docs.