Anyone configuring zeroclaw with fallback config?

[leonardorey1992]

Hi there is just install zeroclaw with docker and it's amazing, love it so far!

But my question is, im using gemini as provider and i know there are several models there that i could work with for most of the cases like gemma ones and so

But at the moment i config it, it only allows me to use one

since free tier gemini is so short in RPM i think it could be great to have a fallback-rotating within the same provider.

It's this possible right now and im missing something? If not i think is the kind of config that for free tiers could be great to have, not like defining agents with an specific version, but to give them a subset so they van rotate

I check the doc about something like this

[reliability]
model_fallbacks = { "claude-opus-4-20250514" = ["claude-sonnet-4-20250514", "gpt-4o"] }

but dunno how to do it properly

thank u for the project and also for the effort :D

so far im trying with

[reliability.model_fallbacks]
gemma-4-26b-a4b-it = [ "gemini-2.5-flash", "gemini-2.5-flash-lite", "gemini-2.5-pro", "gemini-2.0-flash", "gemini-2.0-flash-lite", "gemma-3-1b-it", "gemma-3-4b-it", "gemma-3-12b-it", "gemma-3-27b-it", "gemma-3-2b-it", "gemma-4-26b-a4b-it", "gemma-4-31b-it", "text-embedding-004", "gemini-3-flash", "gemini-3.1-flash-lite", "gemini-3.1-pro", "text-embedding-005", "gemini-robotics-er-1.5-preview" ]

But not sure is the way...

[Audacity88]

@leonardorey1992, the old [reliability.model_fallbacks] shape is not the current config path.

Current fallback lives on provider aliases:

• fallback_models tries alternate model IDs on the same provider alias.
• fallback tries other configured provider aliases.

For the same Gemini endpoint/key, use fallback_models:

[providers.models.gemini.main]
model = "gemini-2.5-flash"
fallback_models = [
  "gemini-2.0-flash",
  "gemini-2.0-flash-lite",
]

[agents.default]
model_provider = "gemini.main"

If you need different endpoints, keys, or provider families, create separate aliases and list them in fallback:

[providers.models.gemini.main]
model = "gemini-2.5-flash"
fallback = ["gemini.backup"]

[providers.models.gemini.backup]
model = "gemini-2.0-flash"

One important boundary: fallback is failure fallback, not round-robin load balancing. It is used after a request fails/retries, for example rate limit, provider down, or model unavailable. If you want deliberate rotation across free-tier models before failure, that should be a separate feature issue.

Docs:

• Provider fallback: https://docs.zeroclawlabs.ai/master/en/providers/configuration.html#fallback-on-failure