Skip to main content
GGA works with any AI provider you already have installed. Set PROVIDER in your .gga config to point to one of the supported options.

​
Provider overview

ProviderConfig valueCLI command usedInstallation
Claudeclaudeecho "prompt" | claude --printclaude.ai/code
Geminigeminigemini -p "prompt"github.com/google-gemini/gemini-cli
Codexcodexcodex exec "prompt"npm i -g @openai/codex
OpenCodeopencode or opencode:<model>opencode run "prompt"opencode.ai
Ollamaollama:<model>ollama run <model> "prompt"ollama.ai
LM Studiolmstudio or lmstudio:<model>HTTP API call to local serverlmstudio.ai
GitHub Modelsgithub:<model>HTTP API via gh auth tokengithub.com/marketplace/models
Override the provider for a single run without changing your config file:
GGA_PROVIDER="gemini" gga run

​
Provider setup

Claude is the recommended provider for strict mode and CI/CD pipelines. It reliably follows structured instructions and consistently outputs the STATUS: PASSED / STATUS: FAILED format.InstallationInstall the Claude Code CLI from claude.ai/code.Verify
echo "Say hello" | claude --print
Config
.gga
PROVIDER="claude"
How GGA calls it
printf '%s' "$prompt" | claude --print
Google’s Gemini CLI. Built into Antigravity IDE — if you use Antigravity, Gemini is already available in your integrated terminal.InstallationInstall the Gemini CLI from github.com/google-gemini/gemini-cli.Verify
echo "Say hello" | gemini
You must be authenticated (gemini login) before GGA can use it.Config
.gga
PROVIDER="gemini"
How GGA calls it
gemini -p "$prompt"
OpenAI’s Codex CLI, optimized for code tasks.Installation
npm i -g @openai/codex
Verify
codex exec "Say hello"
Config
.gga
PROVIDER="codex"
How GGA calls it
codex exec "$prompt"
OpenCode is a provider-agnostic AI coding CLI. You can use it with its default model or pin a specific model.InstallationInstall from opencode.ai.Config
PROVIDER="opencode"
How GGA calls it
# Default model
opencode run "$prompt"

# Specific model
opencode run --model "anthropic/claude-opus-4-5" "$prompt"
Run models locally with Ollama. No API keys required and all data stays on your machine.InstallationInstall from ollama.ai or via Homebrew:
brew install ollama
Pull a model
ollama pull llama3.2
ollama pull codellama
ollama pull qwen2.5-coder
ollama pull deepseek-coder
Config
PROVIDER="ollama:llama3.2"
A model name is required — PROVIDER="ollama" without a model name will fail validation.Custom hostIf your Ollama instance is not running on localhost:11434, set OLLAMA_HOST:
.gga
OLLAMA_HOST="http://192.168.1.100:11434"
PROVIDER="ollama:llama3.2"
How GGA calls itGGA prefers the Ollama REST API (via curl + python3) for clean output. It falls back to the ollama run CLI if those tools are not available.
# API path (preferred)
curl -s http://localhost:11434/api/generate \
  -d '{"model":"llama3.2","prompt":"...","stream":false}'

# CLI fallback
ollama run llama3.2 "$prompt"
Ollama is a pure LLM without file-reading tools. If your AGENTS.md uses references to other files (e.g. - UI guidelines: ui/AGENTS.md), Ollama cannot read them. Consolidate all rules into a single file when using Ollama.
LM Studio runs models locally through an OpenAI-compatible HTTP API. No API keys required.Setup
1

Download LM Studio

Install from lmstudio.ai.
2

Download a model

Use the model browser inside LM Studio to download a model.
3

Start the local server

Open the Local Server tab in LM Studio and start the server. It listens on http://localhost:1234/v1 by default.
4

Configure GGA

Set PROVIDER in your .gga file.
Config
PROVIDER="lmstudio"
Custom hostIf your LM Studio server is on a different host or port, set LMSTUDIO_HOST:
.gga
LMSTUDIO_HOST="http://localhost:8080/v1"
PROVIDER="lmstudio"
Verify the connection
curl http://localhost:1234/v1/models
How GGA calls itGGA calls the OpenAI-compatible chat/completions endpoint directly:
curl -s http://localhost:1234/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"local-model","messages":[{"role":"user","content":"..."}],"stream":false}'
curl and python3 must be available on your system. GGA uses python3 for safe JSON parsing; a sed/grep fallback is used if python3 is not available.
Access dozens of hosted models — GPT-4o, GPT-4.1, DeepSeek R1, Grok 3, Phi-4, LLaMA, and more — using your existing GitHub account. No separate API keys needed.Setup
1

Install GitHub CLI

brew install gh
# or see https://cli.github.com
2

Authenticate

gh auth login
3

Configure GGA

Set PROVIDER with your chosen model.
Config
PROVIDER="github:gpt-4o"
A model name is required — PROVIDER="github" without a model will fail validation.Browse all available models at github.com/marketplace/models.How GGA calls itGGA retrieves a token via gh auth token and calls the GitHub Models inference endpoint:
curl -sS https://models.inference.ai.azure.com/chat/completions \
  -H "Authorization: Bearer $(gh auth token)" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-4o","messages":[...],"temperature":0.2}'
Both gh CLI and curl must be installed and available in PATH.

Build docs developers (and LLMs) love

Get started for freeTalk to us