LLM & embedding models

Providers

• OpenRouter — Unified gateway to Claude, Gemini, GPT (default)
• Ollama — Local LLM inference

LLM aliases

Alias	Model ID
gemini-flash	google/gemini-3-flash-preview
claude-sonnet	anthropic/claude-sonnet-4.5
gpt-pro	openai/gpt-5.2-pro

Configure via shipspec.json (llm.provider, llm.modelName) or ship-spec model set.

Embeddings

Provider	Default Model	Dimensions
OpenRouter	mistralai/codestral-embed-2505	Auto
Ollama	nomic-embed-text	768

Token budgeting:

• maxContextTokens: 16000
• reservedOutputTokens: 4000
• RAG allocation ~70% of available context; low-relevance chunks pruned to fit.