token limit per model config entry?

[rocky cosmos]

hey is it possible setup new model usage so that openclaw will first use 1m tokens of gpt-5.1-codex, then when that is exhausted immediately switch to 10 mill tokens of 5.1-codex-mini and then when that is used switch to the third method? (which will probably be gemini cli) and then have those 1m and 10m token allowances reset daily?

[lean whale]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[solid night]

Not as a first‑class “token budget router”, no. OpenClaw can fail over on errors (429/quota/rate limit/auth/timeouts), but it doesn’t have a config like “spend 1M tokens on model A, then automatically switch to B until 10M, reset daily”.

What you can do (and it gets you pretty close):

1) Chain models with failover (error-driven)

Configure your primary + ordered fallbacks:

• agents.defaults.model.primary
• agents.defaults.model.fallbacks

Config reference: https://docs.openclaw.ai/gateway/configuration-reference (see agents.defaults.model)
Failover behavior: https://docs.openclaw.ai/help/faq#model-failover-and-all-models-failed

2) Enforce the “1M then 10M” outside OpenClaw

If you can put hard daily limits on the upstream credentials (provider-side limits, or a proxy), then once model A hits its quota it will start throwing 429/quota → OpenClaw falls back to model B automatically.

One common way is LiteLLM virtual keys with budgets (budgeting is in $ not tokens, but you can approximate):