#Api rate limit

1 messages · Page 1 of 1 (latest)

hushed kelp
#

Hi whenever I try to write something to openclaw chat it says api rate limit is reached even when opening a new session what do I need to do, to fix it?

spice treeBOT
#

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

mild etherBOT
#

That message almost always means your model provider (Anthropic/OpenAI/OpenRouter/etc) is returning HTTP 429 / quota exceeded. Starting a new OpenClaw session won’t fix it, because the limit is tied to your API key / setup-token, not the chat session.

To diagnose quickly, can you paste:

  1. openclaw --version
  2. The exact error text you see (if it mentions HTTP 429, rate_limit_error, insufficient_quota, etc)
  3. openclaw models status (safe to share)
  4. If you can: openclaw status --all (it redacts secrets)

Common fixes depending on provider:

  • Anthropic (Claude) 429: you’ve exhausted your rate/quota window → wait for the window to reset or upgrade/raise limits. If you’re using a Claude subscription setup-token, it’s especially easy to hit limits. Docs: https://docs.openclaw.ai/help/faq#why-am-i-seeing-http-429-ratelimiterror-from-anthropic
  • OpenAI: usually billing/quota (or RPM/TPM) → check OpenAI usage/limits for that key.
  • If it’s happening “instantly” every time: you may have a background loop (cron/heartbeat/agent stuck retrying) burning requests—openclaw logs --follow will usually show that.

Also: set up a fallback model so OpenClaw can keep replying when the primary provider is rate-limited (docs mention this in the same FAQ + troubleshooting): https://docs.openclaw.ai/gateway/troubleshooting

#

Drop the outputs above and I’ll tell you exactly which case you’re in.

hushed kelp
#

I use openai, but codex

mild etherBOT
#

Got it — if you’re on Codex (OpenAI subscription OAuth), that “API rate limit reached” is almost certainly OpenAI’s Codex-side rate limit, and a new OpenClaw session won’t reset it.

A few fast checks to pinpoint the fix:

  1. What does this show?
openclaw models status

I’m looking for whether your default is openai-codex:... (OAuth/subscription) vs openai:... (API key).

  1. Grab the real underlying 429 from logs (OpenClaw often shows a generic message in chat):
openclaw logs --follow

Then send one message, and paste the few log lines around the error (it’ll usually say 429 + a reason like rate_limit_exceeded / insufficient_quota / etc).

Most common outcomes

#
  • If you’re on openai-codex (OAuth): limits can be tight. Fix is usually: wait for the window to reset (often 1–10 min), and make sure you don’t have a background loop (cron/heartbeat) hammering requests.
  • If you want it to stop happening: switch your default to the regular openai provider with an API key (Codex OAuth is convenient, but not always the most reliable for sustained chat throughput). Also set a fallback provider so you still get replies when Codex is throttled.

Post openclaw --version + openclaw models status + the log snippet and I’ll tell you exactly which one you’re hitting and what to change.

hushed kelp
#

OAuth/token status

  • openai-codex usage: 5h 81% left ⏱2h 37m · Week 0% left ⏱1d 19h
    • openai-codex:default ok expires in 7d
      this is what I see on my temrinal
#

How do I increase it

mild etherBOT
#

That output explains it:

  • openai-codex usage … Week 0% left ⏱1d 19h = your Codex OAuth weekly allowance is fully exhausted.
  • So OpenClaw will keep hitting rate limit / quota no matter how many new sessions you start, until that weekly timer resets (in ~1 day 19 hours).

Your options

  1. Wait for the weekly window to reset → then Codex will work again automatically.

  2. Switch to the normal OpenAI API-key provider (usage-based billing) so you can keep chatting now:

  • Run the wizard just for OpenAI API key auth:
    openclaw onboard --auth-choice openai-api-key
# or non-interactive:
openclaw onboard --openai-api-key "$OPENAI_API_KEY"
  • Then set your agent’s default model/provider to the openai/* (API key) path (instead of openai-codex/*).
#

Docs for that flow: https://docs.openclaw.ai/providers/openai (see “Option A: OpenAI API key”)

If you paste openclaw models status (the full output), I can tell you exactly what to change so OpenClaw prefers the API-key path until Codex weekly resets.

#

You can’t “turn up” that Codex OAuth weekly quota from inside OpenClaw — that Week 0% limit is coming from your OpenAI Codex/ChatGPT OAuth account, and OpenClaw is just reporting/enforcing what the provider allows.

To get more capacity, your real options are:

  1. Upgrade whatever Codex/ChatGPT plan you’re on (if OpenAI offers a higher Codex allowance for your account), then re-check openclaw models status.

  2. Switch to an OpenAI API key (usage-based billing) instead of openai-codex OAuth. That’s the path where you can usually get higher limits by:

    • adding billing / increasing spend cap, and
    • requesting/receiving higher rate limits on the OpenAI Platform.

    OpenClaw setup docs: https://docs.openclaw.ai/providers/openai (Option A: OpenAI API key)

  3. Reduce usage so the weekly bucket lasts longer (shorter prompts, fewer long-context chats, avoid huge pasted logs, etc.) — but it won’t help once the weekly bucket is already at 0%.

If you tell me whether you prefer (A) “stay on Codex OAuth” or (B) “move to API key”, I can give you the exact OpenClaw steps for that path.

hushed kelp
#

Alright so I pay for chatgpt monthly and what I used within that week is fully used so I need custom api key for billing to talk more

mild etherBOT
#

Yep — that’s exactly it.

A ChatGPT monthly subscription (Codex OAuth) has its own weekly allowance, and once that “Week” bucket hits 0%, OpenClaw can’t use Codex again until it resets.

If you want to keep chatting beyond that, you need to switch OpenClaw to the OpenAI Platform API (custom API key + pay‑as‑you‑go billing).

What to do

  1. Create an OpenAI API key on the OpenAI Platform (separate from ChatGPT)

    • Make sure billing is enabled there (otherwise you’ll hit quota errors fast).

  2. On your OpenClaw machine, run the wizard for OpenAI API key auth:

openclaw onboard --auth-choice openai-api-key

(or non-interactive)