#OpenClaw rate limit error

1 messages · Page 1 of 1 (latest)

crude kiln
#

My agents keep failing with ⚠️ API rate limit reached. Please try again later. errors even though:

My claude.ai account works fine in the browser
The OAuth token type (sk-ant-oat01-) tests successfully directly against the Anthropic API
console.anthropic.com shows zero API usage this month

Running OpenClaw 2026.2.23 on Mac, using mode: token (claude.ai OAuth). Gateway restarts don't fix it. OpenClaw puts itself in a self-imposed cooldown after 3 failures, but clearing that doesn't help either — it just hits the same error again immediately.
Is this a known issue with claude.ai OAuth tokens getting rate limited separately from normal browser usage? Is the fix to switch to a paid API key?

vast berryBOT
#

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

visual jungleBOT
#

Yep — this is (unfortunately) expected behavior with Claude subscription auth (setup-token), and it won’t correlate with what you see in console.anthropic.com.

Key points:

  • sk-ant-oat01-… is what OpenClaw calls an Anthropic setup-token (subscription auth). That usage does not show up in the Anthropic Console usage dashboard, which is primarily for API-key / console billing traffic. So “0 usage this month” there can still be consistent with getting 429 rate_limit_error via the setup-token.
  • Claude-in-browser working also doesn’t guarantee the same quota/limits for API-style usage. The subscription token path can be rate-limited separately (different request patterns, concurrency, model access, etc.).

What I’d do next (to avoid guessing):

  1. Confirm the exact 429 reason in logs (OpenClaw sometimes summarizes different 429-ish failures into the same “rate limit reached” UI message):
openclaw status --all
# or at least
openclaw logs --follow

If you paste the snippet around the Anthropic failure, it usually includes the real message / request id (tokens redacted).

#
  1. Check whether you’re hitting the long-context gate:
  1. If it’s a plain rate_limit_error with no extra text: switching to a paid Anthropic API key is the pragmatic fix (and the recommended “production-ish” path):
    https://docs.openclaw.ai/providers/anthropic
    You can keep the setup-token as a fallback if you want, but API key billing is much more predictable.

If you drop the openclaw status --all (or just the 429 chunk from openclaw logs --follow), I can tell you which branch it is (long-context gate vs real rate-limit window vs something being misclassified).

crude kiln
#

"Does anyone know how much OAuth usage you get on the Claude Max $200 plan? And how long does the cooldown last when you hit the rate limit?"