LLM Providers

How to configure skene with different LLM providers, including cloud APIs and local models.

Provider comparison

Setting the provider

There are three ways to configure your provider, model, and API key:

# 1. CLI flags (highest priority)
uvx skene analyze . --provider gemini --model gemini-3-flash-preview --api-key "your-key"

# 2. Environment variables
export SKENE_API_KEY="your-key"
export SKENE_PROVIDER="gemini"

# 3. Config file (.skene.config)
uvx skene config  # Interactive setup

See Configuration for the full priority order.

OpenAI

The default provider. Get an API key at platform.openai.com/api-keys.

Any OpenAI model can be used via --model. The default is gpt-4o.

uvx skene analyze . --provider openai --api-key "sk-..."

# gpt-4o is the default, but you can specify any OpenAI model
uvx skene analyze . --model gpt-4o-mini --api-key "sk-..."

Gemini

Google's Gemini models via the v1beta API. Get an API key at aistudio.google.com/apikey.

Any Gemini model can be used via --model. The default is gemini-3-flash-preview.

uvx skene analyze . --provider gemini --api-key "your-gemini-key"

# Use a specific model
uvx skene analyze . --provider gemini --model gemini-2.5-pro --api-key "your-gemini-key"

Note: The v1beta API requires the -preview suffix on Gemini 3.x models.

Anthropic / Claude

Anthropic's Claude models. Get an API key at console.anthropic.com. Both anthropic and claude work as provider names.

Any Claude model can be used via --model. The default is claude-sonnet-4-5.

uvx skene analyze . --provider anthropic --api-key "sk-ant-..."

# Or use the "claude" alias
uvx skene analyze . --provider claude --api-key "sk-ant-..."

# Use a specific model
uvx skene analyze . --provider claude --model claude-haiku-4-5 --api-key "sk-ant-..."

LM Studio

Run models locally with LM Studio. No API key required.

Use --model to specify whichever model you have loaded in LM Studio. If omitted, skene sends custom-model as the model name (LM Studio typically ignores this and uses whichever model is currently loaded).

# Make sure LM Studio is running with a model loaded
uvx skene analyze . --provider lmstudio

# Specify the model name if needed
uvx skene analyze . --provider lmstudio --model "your-loaded-model"

Default server URL: http://localhost:1234/v1

To use a custom port, set the LMSTUDIO_BASE_URL environment variable:

export LMSTUDIO_BASE_URL="http://localhost:8080/v1"

The provider also accepts lm-studio and lm_studio as aliases.

See Troubleshooting for common LM Studio issues.

Ollama

Run models locally with Ollama. No API key required.

Use --model to specify whichever model you have pulled in Ollama. The default is llama3.3.

# Pull a model first
ollama pull llama3.3

# Make sure Ollama is running
ollama serve

# Analyze
uvx skene analyze . --provider ollama

# Specify a model
uvx skene analyze . --provider ollama --model mistral

Default server URL: http://localhost:11434/v1

To use a custom port, set the OLLAMA_BASE_URL environment variable:

export OLLAMA_BASE_URL="http://localhost:8080/v1"

See Troubleshooting for common Ollama issues.

Generic (OpenAI-compatible)

Connect to any OpenAI-compatible API endpoint. Requires --base-url or the SKENE_BASE_URL environment variable.

# With API key
uvx skene analyze . --provider generic --base-url "https://your-api.com/v1" --api-key "your-key" --model "your-model"

# Local endpoint without API key
uvx skene analyze . --provider generic --base-url "http://localhost:8000/v1" --model "local-model"

The provider also accepts openai-compatible and openai_compatible as aliases.

Rate limiting & fallback

When an LLM provider returns a rate limit error, skene automatically falls back to a cheaper model to keep the workflow moving. This is convenient for interactive use but can corrupt results during benchmarking or when you need guaranteed output from a specific model.

Disabling fallback

Pass --no-fallback to disable model switching. Instead of falling back, the CLI retries the same model with exponential backoff and raises an error if all retries are exhausted:

uvx skene analyze . --provider gemini --model gemini-3-flash-preview --no-fallback
uvx skene plan --no-fallback
uvx skene build --no-fallback

This flag is available on the analyze, plan, and build commands.

Next steps

• Configuration — Save provider settings to a config file
• Troubleshooting — Fix common provider issues