LLM Providers

Overview

Wilson supports 9 LLM providers. Provider resolution is prefix-based — the model name prefix determines which provider handles the request.

Provider Table

Provider	Prefix	API Key Env Var	Example Model
OpenAI	`gpt-`	`OPENAI_API_KEY`	`gpt-4o`
Anthropic	`claude-`	`ANTHROPIC_API_KEY`	`claude-sonnet-4-20250514`
Google	`gemini-`	`GOOGLE_API_KEY`	`gemini-2.0-flash`
xAI	`grok-`	`XAI_API_KEY`	`grok-2`
Moonshot	`kimi-`	`MOONSHOT_API_KEY`	`kimi-chat`
DeepSeek	`deepseek-`	`DEEPSEEK_API_KEY`	`deepseek-chat`
OpenRouter	`openrouter:`	`OPENROUTER_API_KEY`	`openrouter:anthropic/claude-3.5-sonnet`
LiteLLM	`litellm:`	`LITELLM_API_KEY`	`litellm:gpt-4o`
Ollama	`ollama:`	None (local)	`ollama:llama3.1`

Switching Providers

Interactive Mode

Use the /model command:

/model ollama:llama3.1
/model gpt-4o
/model claude-sonnet-4-20250514

Or just /model to open the model selector.

Headless Mode

Set the DEFAULT_MODEL environment variable in .env:

DEFAULT_MODEL=ollama:llama3.1

Provider Details

Ollama (Local)

The privacy-first option. Runs entirely on your machine with no API calls.

# Install
brew install ollama

# Start server
ollama serve

# Pull a model
ollama pull llama3.1

No API key needed. Set OLLAMA_BASE_URL if running on a non-default port:

OLLAMA_BASE_URL=http://localhost:11434

Recommended models for financial tasks:

Model	Size	Quality	Speed
`llama3.1:70b`	40GB	Best	Slow
`llama3.1:8b`	4.7GB	Good	Fast
`mistral`	4.1GB	Good	Fast
`deepseek-coder-v2`	8.9GB	Good for structured data	Medium

OpenAI

OPENAI_API_KEY=sk-...

Best overall quality for complex financial reasoning. gpt-4o recommended.

Anthropic

ANTHROPIC_API_KEY=sk-ant-...

Strong for nuanced analysis and following detailed skill instructions.

Google

GOOGLE_API_KEY=AIza...

gemini-2.0-flash offers good quality at low cost.

OpenRouter

Access 100+ models through a single API key:

OPENROUTER_API_KEY=sk-or-...

Prefix models with openrouter: — e.g., openrouter:anthropic/claude-3.5-sonnet.

LiteLLM

Proxy server that provides a unified interface to 100+ LLM providers:

LITELLM_API_KEY=sk-...
LITELLM_BASE_URL=http://localhost:4000

Architecture

Wilson uses a central callLlm() facade with retry logic that routes to provider-specific adapters. Each adapter normalizes the provider’s API into Wilson’s internal message format.

Providers that support the OpenAI-compatible API (xAI, Moonshot, DeepSeek, OpenRouter, LiteLLM) share a single adapter implementation, reducing maintenance overhead.