Skip to content

config.md

Latest commit

 

History

History
130 lines (112 loc) · 4.51 KB

config.md

File metadata and controls

130 lines (112 loc) · 4.51 KB

Configuration Reference

Config file

~/.free-coding-models.json is created automatically on first run (permissions 0600).

{
  "apiKeys": {
    "nvidia":      "nvapi-xxx",
    "groq":        "gsk_xxx",
    "cerebras":    "csk_xxx",
    "openrouter":  "sk-or-xxx",
    "huggingface": "hf_xxx",
    "replicate":   "r8_xxx",
    "deepinfra":   "di_xxx",
    "siliconflow": "sk_xxx",
    "together":    "together_xxx",
    "cloudflare":  "cf_xxx",
    "perplexity":  "pplx_xxx",
    "zai":         "zai-xxx"
  },
  "providers": {
    "nvidia":      { "enabled": true },
    "groq":        { "enabled": true },
    "cerebras":    { "enabled": true },
    "openrouter":  { "enabled": true },
    "huggingface": { "enabled": true },
    "replicate":   { "enabled": true },
    "deepinfra":   { "enabled": true },
    "siliconflow": { "enabled": true },
    "together":    { "enabled": true },
    "cloudflare":  { "enabled": true },
    "perplexity":  { "enabled": true },
    "zai":         { "enabled": true }
  },
  "settings": {
    "hideUnconfiguredModels": true
  },
  "router": {
    "enabled": true,
    "activeSet": "fast-coding",
    "port": 19280,
    "probeMode": "balanced",
    "sets": {
      "fast-coding": {
        "name": "fast-coding",
        "models": [
          { "provider": "groq", "model": "openai/gpt-oss-120b", "priority": 1 },
          { "provider": "cerebras", "model": "gpt-oss-120b", "priority": 2 }
        ],
        "created": "2026-04-22T10:00:00.000Z"
      }
    }
  },
  "favorites": [
    "nvidia/deepseek-ai/deepseek-v3.2"
  ]
}

Environment variables

The TUI uses env vars before config values. The Smart Model Router daemon uses config keys first, then env vars as a fallback, because background services may not inherit your shell environment.

Variable Provider
NVIDIA_API_KEY NVIDIA NIM
GROQ_API_KEY Groq
CEREBRAS_API_KEY Cerebras
SAMBANOVA_API_KEY SambaNova
OPENROUTER_API_KEY OpenRouter
HUGGINGFACE_API_KEY / HF_TOKEN Hugging Face
REPLICATE_API_TOKEN Replicate
DEEPINFRA_API_KEY / DEEPINFRA_TOKEN DeepInfra
CODESTRAL_API_KEY Mistral Codestral
HYPERBOLIC_API_KEY Hyperbolic
SCALEWAY_API_KEY Scaleway
GOOGLE_API_KEY Google AI Studio
SILICONFLOW_API_KEY SiliconFlow
TOGETHER_API_KEY Together AI
CLOUDFLARE_API_TOKEN / CLOUDFLARE_API_KEY Cloudflare Workers AI
CLOUDFLARE_ACCOUNT_ID Cloudflare (required for endpoint URL)
PERPLEXITY_API_KEY / PPLX_API_KEY Perplexity
ZAI_API_KEY ZAI
DASHSCOPE_API_KEY Alibaba Cloud (DashScope)

Runtime settings

Setting Default Description
Ping timeout 15 s Per-request timeout. Slow models get more time.
Ping cadence 2 s → 10 s → 30 s Fast burst at startup, then normal, then idle slowdown
Configured-only on Only show providers with API keys. Toggle with E.
Favorites persistent Stored in config file, survive app restarts and updates

Smart Model Router

The router section is created automatically when you run free-coding-models --daemon-bg or free-coding-models --daemon. It controls the local OpenAI-compatible daemon.

Field Default Description
enabled false until started Whether router config is active. Starting the daemon enables it.
port 19280 Preferred localhost port. If occupied, the daemon scans 19280 through 19289.
activeSet fast-coding Model set used by /v1/chat/completions.
probeMode balanced Health probe intensity: eco, balanced, or aggressive.
sets auto-created Named ordered model groups. Lower priority numbers are tried first during cold start.

Each set model entry uses:

Field Description
provider Provider key from sources.js, such as groq or nvidia.
model Provider-native model ID.
priority User priority inside the set. The config normalizer keeps priorities contiguous.

Runtime files:

File Purpose
~/.free-coding-models-daemon.pid Running daemon PID for stop/status discovery.
~/.free-coding-models-daemon.port Actual port selected after fallback scanning.
~/.free-coding-models-daemon.log Rotating daemon lifecycle, probe, and routing metadata logs.
~/.free-coding-models-tokens.json Daily and all-time token counters from successful non-streaming responses.