Model providers

StockSage uses LiteLLM via LLM_MODEL in provider/model-name form. Keys are standard LiteLLM env vars.

Run python -m src.core.config.check after editing .env to catch missing keys or unreachable Ollama.

Ollama (local, no API key)

1. Install Ollama and start it.
2. Pull a model, e.g. ollama pull qwen2.5:14b-instruct.
3. In .env:

LLM_MODEL=ollama/qwen2.5:14b-instruct
OLLAMA_BASE_URL=http://localhost:11434

Optional fallback when Ollama is down (requires that provider’s key):

LLM_FALLBACK_MODEL=openai/gpt-4o-mini
OPENAI_API_KEY=sk-...

OpenAI

LLM_MODEL=openai/gpt-4o-mini
OPENAI_API_KEY=sk-...

DeepSeek

LLM_MODEL=deepseek/deepseek-chat
DEEPSEEK_API_KEY=...

Google Gemini

LLM_MODEL=gemini/gemini-2.5-flash
GEMINI_API_KEY=...

Groq

LLM_MODEL=groq/llama-3.3-70b-versatile
GROQ_API_KEY=...

Anthropic

LLM_MODEL=anthropic/claude-3-5-haiku-20241022
ANTHROPIC_API_KEY=...

Serper (news search tool)

Optional but improves live news context:

SERPER_API_KEY=...

Adding a new provider

1. Confirm LiteLLM supports it: providers.
2. Set LLM_MODEL=provider/model-id.
3. Set the API key env var LiteLLM expects for that provider.
4. If you want check to validate keys, add a mapping in src/core/config/check.py (_PROVIDER_ENV_KEYS).