Use CoralBricks with Pi

Run the Pi coding agent on CoralBricks open models — GLM and Kimi with up to 1M context — with one provider block in models.json.

Pi is a minimal, extensible terminal coding harness. It takes custom OpenAI-compatible providers via a single JSON file, so you can run its agent loop on GLM 5.3, Kimi K3, or GPT-OSS 120B through CoralBricks.

1. Install Pi

npm install -g --ignore-scripts @earendil-works/pi-coding-agent

2. Get an API key

Create a key from your API keys page (format cb_...), then export it:

export CORALBRICKS_API_KEY="cb_your_key_here"

Add the line to your shell profile so it persists. (Alternatively, store the key with Pi's /login command instead of using an env var.)

3. Add CoralBricks as a provider

Create (or extend) ~/.pi/agent/models.json:

{
  "providers": {
    "coralbricks": {
      "name": "CoralBricks",
      "baseUrl": "https://inference.coralbricks.ai/v1",
      "api": "openai-completions",
      "apiKey": "$CORALBRICKS_API_KEY",
      "models": [
        {
          "id": "glm-5.3-fp4",
          "name": "GLM 5.3",
          "reasoning": true,
          "input": ["text"],
          "contextWindow": 1048576,
          "maxTokens": 32768,
          "cost": { "input": 1.12, "output": 4.4, "cacheRead": 0, "cacheWrite": 0 }
        },
        {
          "id": "kimi-k3",
          "name": "Kimi K3",
          "reasoning": true,
          "input": ["text", "image"],
          "contextWindow": 1048576,
          "maxTokens": 32768,
          "cost": { "input": 3, "output": 15, "cacheRead": 0, "cacheWrite": 0 }
        },
        {
          "id": "gpt-oss-120b",
          "name": "GPT-OSS 120B",
          "reasoning": true,
          "input": ["text"],
          "contextWindow": 131072,
          "maxTokens": 32768,
          "cost": { "input": 0.12, "output": 0.6, "cacheRead": 0, "cacheWrite": 0 }
        }
      ]
    }
  }
}

The file reloads whenever you open /model — no restart needed.

4. Run

pi --model coralbricks/glm-5.3-fp4

Or pick the model interactively with /model. One-shot mode works too:

pi -p --model coralbricks/glm-5.3-fp4 "explain this repo"

Available models

Model	Model ID	Context	Input $/M	Output $/M
GLM 5.3	`glm-5.3-fp4`	1M	$1.12	$4.40
Kimi K3	`kimi-k3`	1M	$3.00	$15.00
GPT-OSS 120B	`gpt-oss-120b`	128K	$0.12	$0.60

Cached input tokens are always free.

Notes

No compatibility flags needed — the CoralBricks gateway accepts the developer role and honors reasoning_effort, so Pi's defaults for reasoning-capable models work as-is (no compat overrides).
Reasoning renders in Pi — GLM 5.3 and Kimi K3 emit reasoning that Pi displays inline; set "reasoning": true as shown.
Long context is the point: GLM 5.3 and Kimi K3 accept up to 1M tokens, so large-repo Pi sessions fit without truncation.
The same key works in OpenCode, Cline, Codex CLI, Cursor, and anything else that speaks OpenAI-compatible providers. See the API reference for the full surface.