Skip to content

integrations.md

Latest commit

 

History

History
128 lines (95 loc) · 3.83 KB

integrations.md

File metadata and controls

128 lines (95 loc) · 3.83 KB

Tool Integrations

Every tool follows the same pattern:

  1. Run free-coding-models --<tool> (or press Z to cycle)
  2. Wait for models to ping (green ✅)
  3. Navigate with ↑↓, press Enter
  4. If the tool CLI is missing, FCM offers a tiny install confirmation and runs the official global install command
  5. FCM writes the selected model into the tool's config and launches it

Tool → Config mapping

Tool Flag Config written
OpenCode CLI --opencode ~/.config/opencode/opencode.json
OpenCode Desktop --opencode-desktop ~/.config/opencode/opencode.json (then opens app)
OpenClaw --openclaw ~/.openclaw/openclaw.json
Crush --crush ~/.config/crush/crush.json
Goose --goose ~/.config/goose/config.yaml + custom_providers/
Aider --aider ~/.aider.conf.yml
Qwen Code --qwen ~/.qwen/settings.json
OpenHands --openhands LLM_MODEL env var
Amp --amp ~/.config/amp/settings.json
Pi --pi ~/.pi/agent/settings.json

OpenCode

free-coding-models --opencode

FCM auto-detects your configured providers, writes the selected model to opencode.json, and launches opencode.

tmux sub-agent panes

When launched inside tmux, FCM auto-adds --port so OpenCode can spawn sub-agent panes:

  • Priority 1: reuse OPENCODE_PORT if valid and free
  • Priority 2: auto-pick first free port in 4096–5095
OPENCODE_PORT=4098 free-coding-models --opencode

ZAI + OpenCode (transparent proxy)

ZAI uses /api/coding/paas/v4/* instead of standard /v1/*. When you pick a ZAI model in OpenCode mode, FCM automatically starts a localhost proxy that rewrites ZAI paths to OpenCode's expected format. It starts on a random port and shuts down when OpenCode exits. No manual config needed.

Manual setup (optional)

Create or edit ~/.config/opencode/opencode.json:

{
  "provider": {
    "nvidia": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "NVIDIA NIM",
      "options": {
        "baseURL": "https://integrate.api.nvidia.com/v1",
        "apiKey": "{env:NVIDIA_API_KEY}"
      }
    }
  },
  "model": "nvidia/deepseek-ai/deepseek-v3.2"
}

⚠️ Free models have usage limits — check build.nvidia.com for quotas.

OpenClaw

free-coding-models --openclaw

FCM writes the selected model as primary into ~/.openclaw/openclaw.json and launches openclaw.

What gets written

{
  "models": {
    "providers": {
      "nvidia": {
        "baseUrl": "https://integrate.api.nvidia.com/v1",
        "api": "openai-completions"
      }
    }
  },
  "env": { "NVIDIA_API_KEY": "nvapi-xxxx" },
  "agents": {
    "defaults": {
      "model": { "primary": "nvidia/deepseek-ai/deepseek-v3.2" },
      "models": { "nvidia/deepseek-ai/deepseek-v3.2": {} }
    }
  }
}

⚠️ providers must be nested under models.providers — a root-level providers key is ignored.

⚠️ The model must also be listed in agents.defaults.models (the allowlist), or OpenClaw rejects it with "not allowed".

Install Endpoints (Y key)

Y opens a step-by-step flow to install a full provider catalog into a tool's config — so you can pick the model inside the tool instead of from FCM.

Steps:

  1. Provider — pick one with a configured API key
  2. Tool — config-based (OpenCode, OpenClaw, Crush, Goose, Pi, Aider, Amp, Qwen) or env-based (OpenHands)
  3. Scope — all models or selected models only
  4. Models (if selected) — multi-select from the provider catalog

Notes:

  • Entries are namespaced under fcm-* in the target config
  • OpenCode CLI and OpenCode Desktop share opencode.json
  • For OpenHands, FCM writes ~/.fcm-openhands-env — source it before launching