Models¶

The models: section in the Runefile configures which LLM providers and models an agent uses.

Structure¶

models:
  providers:
    - anthropic
    - openai
  model_mapping:
    default: claude-sonnet-4-6
    fast: claude-haiku-4-5-20251001
  fallback_policy: next_provider
  token_budget: 100000
  safety_policy: standard

Providers¶

List providers in order of preference. Rune tries each in order when fallback_policy: next_provider.

Provider ID	API Key Env Var	Description
`anthropic`	`ANTHROPIC_API_KEY`	Claude models (Haiku, Sonnet, Opus)
`openai`	`OPENAI_API_KEY`	GPT-4o, GPT-4-turbo, o1, etc.
`gemini`	`GEMINI_API_KEY`	Google Gemini (default model: `gemini-2.0-flash`)
`copilot`	`GITHUB_TOKEN` or `COPILOT_TOKEN`	GitHub Copilot (OpenAI-compatible endpoint)
`claude-code`	`CLAUDE_API_KEY`	Alternative Anthropic key (Claude via `claude-code` alias)

Auto-detection order when env vars are present: anthropic → claude-code → gemini → copilot → openai

Model Mapping¶

Maps logical aliases (referenced by default_model) to physical model IDs:

model_mapping:
  default: claude-sonnet-4-6
  fast: claude-haiku-4-5-20251001
  powerful: claude-opus-4-6

Common model IDs:

Provider	Model IDs
Anthropic	`claude-sonnet-4-6`, `claude-haiku-4-5-20251001`, `claude-opus-4-6`
OpenAI	`gpt-4o`, `gpt-4o-mini`, `gpt-4-turbo`, `o1-preview`
Gemini	`gemini-2.0-flash`, `gemini-1.5-pro`, `gemini-2.0-pro`
Copilot	`gpt-4o` (default for the Copilot endpoint)

Fallback Policy¶

Value	Behaviour
`next_provider`	On failure, try the next provider in the `providers` list
`fail`	Fail immediately on the first provider error

Token Budget¶

Maximum tokens (input + output) per request. Default: 100000.

Safety Policy¶

Value	Description
`standard`	Default safety filters
`strict`	Stricter content filtering
`none`	No filtering