How to Configure Model Fallback in nanobot

Model fallback lets nanobot try a primary model first, then fall back to one or more named presets when the primary provider fails or rate-limits.

What you will build

two or more modelPresets
a primary agents.defaults.modelPreset
an ordered agents.defaults.fallbackModels chain

When to use this

Use fallback when you want better reliability across rate limits, provider outages, local model downtime, or cost-sensitive routing.

Install

python -m pip install nanobot-ai
nanobot onboard --wizard
nanobot agent -m "Hello!"

Verify each provider works before adding it as a fallback.

Minimal working example

Merge this shape into ~/.nanobot/config.json and replace provider/model names with ones you control:

{
  "modelPresets": {
    "fast": {
      "label": "Fast",
      "provider": "openrouter",
      "model": "anthropic/claude-sonnet-4.5",
      "maxTokens": 4096,
      "contextWindowTokens": 65536,
      "temperature": 0.1
    },
    "deep": {
      "label": "Deep",
      "provider": "anthropic",
      "model": "claude-sonnet-4-5",
      "maxTokens": 4096,
      "contextWindowTokens": 200000,
      "temperature": 0.1
    }
  },
  "agents": {
    "defaults": {
      "modelPreset": "fast",
      "fallbackModels": ["deep"]
    }
  }
}

String entries in fallbackModels are preset names, not raw model IDs.

Production notes

Keep fallback context windows realistic; smaller fallback windows constrain how much context can fit.
Put cheaper or faster fallbacks before expensive ones when acceptable.
Use /model <preset> for runtime switching without editing config.
Keep labels human-readable for WebUI model lists.

Security notes

Different providers may have different data handling policies.
Do not put provider keys directly in shared config files.
Confirm fallback models can safely receive the same prompts and files.

Troubleshooting

If a fallback never triggers, confirm the primary error is treated as retryable/fallbackable.
If startup fails, check that each fallback string matches a key under modelPresets.
If output is truncated after fallback, review maxTokens and contextWindowTokens.

2.4 KiB Raw Blame History