AI Providers

June 30, 2026 · View on GitHub

CommitSage supports 10 AI providers — 6 cloud, 1 aggregator, 1 local, plus a generic OpenAI-compatible adapter that closes any self-hosted endpoint. This guide covers setup, features, and trade-offs.

Provider Comparison

Provider	Cost	Runs Locally	API Key	Region notes (RU)
Gemini	Free tier	No	Yes	VPN required
OpenRouter	Free + paid	No	Yes	Works
Groq	Free tier	No	Yes	Works
Anthropic	Paid	No	Yes	VPN required
OpenAI	Paid	No	Yes	Blocked, VPN required
DeepSeek	Paid (cheap)	No	Yes	Works without VPN
xAI	Paid	No	Yes	VPN required
Codestral	Free	No	Yes	Works
Ollama	Free	Yes	No	Works (local)
Custom	Depends	Depends	Optional	Depends

Gemini (Default)

Google's AI models via Google AI Studio.

Setup:

Get a free API key from Google AI Studio
Sidebar → pick Gemini → Set → paste key — or Command Palette → Commit Sage: Set Gemini API Key

Models: auto (default) — fetches available models and tries each until one succeeds. Robust to model deprecations.

Settings:

{
  "commitSage.provider.type": "gemini",
  "commitSage.gemini.model": "auto"
}

OpenRouter

One API key, 300+ models from OpenAI, Anthropic, Google, xAI, DeepSeek, Meta, Mistral, Qwen, and more.

Setup (one-click sign-in):

Sidebar → OpenRouter → Sign in with OpenRouter
Authorize in the browser; the key is created and stored automatically (OAuth PKCE).

Setup (manual key):

Sign up at openrouter.ai/keys
Sidebar → OpenRouter → Set

Free tier: 25+ models with :free suffix (Llama 3.3 70B, DeepSeek V3, GPT-OSS 120B, Qwen3 Coder, etc.). 20 RPM / 50 RPD at $0 balance; 1000 RPD after a $10 top-up.

Free models filter: commitSage.openrouter.preferFreeModels (default true) restricts the model dropdown to free models only. Toggle off to browse all 300+.

Settings:

{
  "commitSage.provider.type": "openrouter",
  "commitSage.openrouter.model": "meta-llama/llama-3.3-70b-instruct:free",
  "commitSage.openrouter.preferFreeModels": true
}

Groq

Fastest commercial inference on the market (~200–500 ms for Llama 3.3 70B).

Setup:

Get a key from console.groq.com/keys (email + phone, no card)
Sidebar → Groq → Set

Free tier: 30 RPM / 1000 RPD on 70B; up to 14 400 RPD on llama-3.1-8b-instant (effectively unlimited for commit messages).

Recommended models: llama-3.1-8b-instant (max throughput), llama-3.3-70b-versatile (quality, default), mixtral-8x7b-32768.

Settings:

{
  "commitSage.provider.type": "groq",
  "commitSage.groq.model": "llama-3.3-70b-versatile"
}

Anthropic Claude

Claude models directly from Anthropic.

Setup:

Get an API key from console.anthropic.com/settings/keys
Sidebar → Anthropic Claude → Set

Models (curated static list — Anthropic has no public /models endpoint):

claude-opus-4-1-20250805 — highest quality
claude-sonnet-4-5-20250929 — default, balanced
claude-haiku-4-5-20251001 — fastest, cheapest
claude-3-5-sonnet-20241022, claude-3-5-haiku-20241022 — previous generation

Keep the list in sync with docs.anthropic.com/en/docs/about-claude/models.

Note: Pro/Max subscription OAuth tokens are not usable here — Anthropic's Usage Policy restricts those to native Anthropic clients. Use an API key from the console.

Settings:

{
  "commitSage.provider.type": "anthropic",
  "commitSage.anthropic.model": "claude-sonnet-4-5-20250929"
}

OpenAI

OpenAI models (GPT-4o, GPT-4.1, o-series).

Setup:

Get an API key from platform.openai.com/api-keys
Sidebar → OpenAI → Set

Custom endpoint: commitSage.openai.baseUrl accepts any OpenAI-compatible URL (Azure OpenAI, LocalAI). For self-hosted setups prefer the dedicated Custom provider instead — it doesn't conflate official OpenAI defaults with self-hosted overrides.

Settings:

{
  "commitSage.provider.type": "openai",
  "commitSage.openai.model": "gpt-4o-mini",
  "commitSage.openai.baseUrl": "https://api.openai.com/v1"
}

DeepSeek

DeepSeek's flagship models. Works without VPN in restricted regions.

Setup:

Get a key from platform.deepseek.com/api_keys
Sidebar → DeepSeek → Set

Models: deepseek-chat (default), deepseek-reasoner.

Settings:

{
  "commitSage.provider.type": "deepseek",
  "commitSage.deepseek.model": "deepseek-chat"
}

xAI Grok

xAI's Grok models.

Setup:

Get a key from console.x.ai
Sidebar → xAI Grok → Set

Models: grok-2-1212 (default, stable), grok-3-mini, others as released.

Settings:

{
  "commitSage.provider.type": "xai",
  "commitSage.xai.model": "grok-2-1212"
}

Codestral

Mistral AI's code-specialized model on the dedicated Codestral subdomain. Free tier.

Setup:

Get a free API key from console.mistral.ai/codestral
Sidebar → Codestral → Set

Models: codestral-latest (default), codestral-2508, codestral-2501, codestral-2405.

Settings:

{
  "commitSage.provider.type": "codestral",
  "commitSage.codestral.model": "codestral-latest"
}

Ollama

Run models locally. No API key, no data leaves your machine.

Setup:

Install Ollama from ollama.com
Pull a model: ollama pull llama3.2 (or qwen2.5-coder:7b for code-specialized)
Sidebar → Ollama (default base URL http://localhost:11434)

Recommended models: qwen2.5-coder:7b (code), llama3.2:3b (fits 4 GB RAM), deepseek-coder-v2:16b (powerful hardware).

Auth token: For hosted Ollama instances behind auth, enable commitSage.ollama.useAuthToken and set the token via Commit Sage: Set Ollama Auth Token.

Settings:

{
  "commitSage.provider.type": "ollama",
  "commitSage.ollama.baseUrl": "http://localhost:11434",
  "commitSage.ollama.model": "llama3.2"
}

Custom OpenAI-compatible

Any OpenAI-compatible chat/completions endpoint — local or remote, with or without auth.

Common targets:

LM Studio — http://localhost:1234/v1
Ollama (OpenAI-compat mode) — http://localhost:11434/v1 (alternative to the native Ollama provider above)
vLLM — http://localhost:8000/v1 (any self-hosted Mistral/Llama/Qwen/etc. served via vLLM)
TGI (HuggingFace Text Generation Inference) — http://your-tgi-host/v1
llama.cpp server — http://localhost:8080/v1
LocalAI — http://localhost:8080/v1
NVIDIA NIM (free hosted models) — https://integrate.api.nvidia.com/v1 + key from https://build.nvidia.com/settings/api-keys; covers free coding models like meta/llama-3.3-70b-instruct, qwen/qwen2.5-coder-32b-instruct
Together AI — https://api.together.xyz/v1 (needs key)
Fireworks — https://api.fireworks.ai/inference/v1 (needs key)
Cerebras, DeepInfra, private deployments — any URL exposing OpenAI wire format

Setup:

Sidebar → Custom (OpenAI-compatible)
Set Base URL to your endpoint
Set Model to whatever model ID your endpoint exposes (free-form text — no listing available)
If your endpoint requires auth: enable Send API key, then Set to paste the key

Settings:

{
  "commitSage.provider.type": "custom",
  "commitSage.custom.baseUrl": "http://localhost:1234/v1",
  "commitSage.custom.model": "qwen2.5-coder",
  "commitSage.custom.useApiKey": false,
  "commitSage.custom.chatCompletionsPath": "/chat/completions"
}

Choosing a Provider

Gemini — best free starting point, auto-mode handles model rotation
OpenRouter — one key, access to almost everything (incl. 25+ free models)
Groq — fastest replies + the most generous free tier
Anthropic — best output quality at higher cost
DeepSeek — works from regions where US providers are blocked
Ollama — best for privacy / offline / sensitive repos
Custom — any self-hosted server (LM Studio, vLLM, llama.cpp) without per-vendor wiring
OpenAI — when you specifically want GPT models
xAI — when you specifically want Grok
Codestral — free code-specialized fallback