pi-opinionated-glm

Provider-only Pi package that registers a dedicated glm provider backed by OpenCode Zen:

• provider name is glm
• base URL is https://opencode.ai/zen/go/v1
• API key env var is OPENCODE_API_KEY
• the package mirrors Pi's built-in OpenCode Go GLM models instead of pretending GLM must come from z.ai directly

That is the point of the package. It is opinionated. For this workspace, OpenCode Go (/zen/go/) is the default GLM backend because the official z.ai/BigModel rate limits are annoying enough to be useless in practice, and the China-optimized gateway offers better latency.

Install

# try temporarily
pi -e ~/coding_agents/pi-opinionated-glm

# or install into the current project
pi install -l ~/coding_agents/pi-opinionated-glm

Set your API key before use:

export OPENCODE_API_KEY=...

What it changes

This package registers a dedicated glm provider that routes requests through OpenCode Zen:

• mirrors Pi's built-in opencode-go GLM chat models at runtime instead of freezing a handwritten duplicate list here
• filters that upstream catalog down to GLM models only, so the package surface stays honest
• seeds a fallback glm-5.1 entry when the local Pi build has not picked it up yet
• relies on Pi's normal OpenAI-compatible transport against https://opencode.ai/zen/go/v1

Registered provider

• Provider name: glm
• API key env var: OPENCODE_API_KEY
• Backend: OpenCode Go (China-optimized Zen gateway)

Usage

Use GLM through the dedicated provider, for example:

• glm/glm-5.1
• glm/glm-5
• glm/glm-4.7

Scope

This package is intentionally narrow:

• it picks an opinionated backend for GLM access
• it does not masquerade as z.ai anymore
• it does not maintain a giant handwritten model registry when Pi already has most of the catalog upstream
• it only adds the small glm-5.1 fallback metadata needed until the local Pi build catches up