Hybrid routing & budgets.

How routing works

Rabbit routes each agent step to a different model based on the step type. Cheap steps (plan, search, review) run locally for free. Only the hard codegen step goes to a frontier cloud model.

routing.yml

Create .rabbit-ide/routing.yml in your project root:

agent:
  plan:       ollama/llama3.1:70b    # local - free
  search:     ollama/qwen2.5-coder   # local - free
  codegen:    anthropic/sonnet-4     # cloud - paid
  review:     openai/gpt-4.1-mini    # cloud - cheap
  fallback:
    - ollama/qwen2.5-coder:32b
    - openai/gpt-4.1-mini
  budget:
    daily_usd:   5.00
    hard_cutoff: true

Fallback chains

When the primary model is unavailable (rate-limited, network error, budget exhausted), the agent walks the fallback chain top-to-bottom until a model responds. A fully local fallback guarantees the agent never stalls.

Budget caps

Set daily_usd to cap cloud spend per day. Whenhard_cutoff: true, the agent stops calling cloud models entirely after the cap is hit. With false, it shows a warning but continues.

Per-project overrides

Each workspace can have its own routing.yml. Global defaults live in ~/.rabbit-ide/routing.yml. Project-level settings override global ones.

Cost tracking

Token usage and estimated cost are shown in the status bar in real time. The metrics tool returns session-level statistics, and wrapped generates period reports (day/week/month).