#gateway

1 messages · Page 1 of 1 (latest)

drowsy eagle
#

18:46:11 Gateway start blocked: set gateway.mode=local (current: unset) or pass --allow-unconfigured.

stone barnBOT
#

That message is the safety guardrail: OpenClaw won’t start a Gateway unless you explicitly say “this machine is allowed to run the Gateway”.

Pick one:

A) You want this machine to run the Gateway (most common)

openclaw config set gateway.mode local
openclaw gateway
# (or if you installed the background service)
openclaw gateway restart

B) You don’t want a local Gateway (you’re connecting to a remote one)

Set remote mode + URL instead (and don’t start the local gateway service):

openclaw config set gateway.mode remote
# 
openclaw --version
openclaw config get gateway.mode

I can tell you which branch you’re currently in.

drowsy eagle
#

what do i do when my browser and gateway have a token mismatch? Still the same token mismatch on my tool side. This is the same issue as before — my tool credentials are cached from the old session.

Open a fresh webchat session in a new tab: http://127.0.0.1:18789/

That'll give me fresh RPC credentials and everything should work. Don't worry, this current chat won't disappear — I'll just have a new connection for tools.

stone barnBOT
#

If the Gateway token changed (or you copied the wrong one into the UI), the browser UI and your running session can keep using the old cached token, so tools keep failing even though the gateway is fine.

Do this in order:

  1. Make sure the Gateway is using the token you think it is
openclaw config get gateway.auth.token
openclaw gateway restart

(If you rotated it, you must restart.)

  1. Fix the browser-side cached token
  • Easiest: open a new tab to the webchat/control UI (your link is fine):
    http://127.0.0.1:18789/
  • If it still “sticks” to the old token: open an incognito/private window, or clear the site data for 127.0.0.1 (localStorage) and reload.
  1. Reconnect in the Control UI with the right auth field
#

In Control → Overview → Gateway Access, there are two boxes side-by-side (Gateway Token vs Password). Make sure you’re filling the one that matches your gateway auth config, then Connect.

Tip: openclaw dashboard usually prints a URL that already includes the token (so you don’t fat-finger it).

If you paste the exact error text you see (and whether you’re using token vs password), I can tell which part is still out of sync.

drowsy eagle
#

it keeps saying API usage limit reached

stone barnBOT
#

“API usage limit reached” isn’t a Gateway-token mismatch thing — it’s almost always your model provider (Anthropic/OpenAI/OpenRouter/etc.) or sometimes the Brave Search API telling you you’ve hit quota/rate limits.

Quick way to tell which:

  • If it happens on any normal message → it’s your LLM provider (billing/quota/rate limit window).
  • If it only happens when the assistant tries to web_search → it’s your Brave Search quota.

Can you paste (with secrets redacted) the first error block from:

openclaw --version
openclaw logs --follow

The logs will say which provider/tool returned the limit.

Things you can do immediately:

  • Provider quota/rate limit: wait for the window to reset, or fix billing/limits on the provider side; optionally set a fallback model/provider so OpenClaw can keep responding when one is cooling down (FAQ mentions this for Anthropic 429s): https://docs.openclaw.ai/start/faq
  • Brave Search quota: add/replace your Brave API key or stop using web_search until it resets: https://docs.openclaw.ai/tools/web
drowsy eagle
#

its definetly anthropic, what can i do to fix this? and what causes this?

stone barnBOT
#

That error means Anthropic is refusing requests before OpenClaw can do anything — you’ve hit an Anthropic quota/usage cap (spend limit / monthly limit) or a short-window rate limit.

What causes it (2 common cases)

  1. Quota / usage cap reached (most likely when the message literally says “API usage limit reached”)
  • Your Anthropic account hit a hard spend limit / monthly budget / free-credit exhaustion.
  • This won’t clear until the billing limit resets or you raise the limit / add funds.
  1. Rate limit (HTTP 429) (bursty usage)
  • Too many requests or too many tokens per minute (big context + tools + retries + multiple sessions/subagents).
  • This usually clears after a short cooldown window.

(FAQ reference for the 429 case: https://docs.openclaw.ai/start/faq)

What you can do to fix it

A) Confirm which one it is (fast)

#

Run:

openclaw models status

If you can, also paste the relevant bit of:

openclaw logs --follow

(redact secrets). The log usually shows 429 vs “insufficient quota / usage limit”.

B) If it’s quota/usage cap

  • Go to the Anthropic Console → billing/usage/limits and raise the hard limit / add payment method / wait for reset.
  • If you’re using a Claude subscription setup-token (not an API key): you may need to wait for the window reset or upgrade (OpenClaw can’t bypass that). Source: https://docs.openclaw.ai/start/faq

C) If it’s rate limit (429)

  • Wait a bit (OpenClaw will often put the profile into a cooldown).
  • Reduce burstiness: avoid running multiple heavy chats/subagents at once; shorten context/tasks.
#

D) Add a fallback so you’re not dead-in-the-water

Set up a second provider/model as a fallback so when Anthropic is limited, OpenClaw can keep replying. (Model config varies by setup; the safest next step is to run openclaw models list / openclaw models set … and configure fallbacks per the docs.)
Docs: https://docs.openclaw.ai/cli/models

E) If you’re on an Anthropic API key, enable/extend prompt caching (reduces token burn)

OpenClaw supports Anthropic prompt caching (API-key only): https://docs.openclaw.ai/providers/anthropic

If you tell me whether you’re using API key or setup-token, and paste the single log line around the failure (redacted), I’ll tell you exactly which bucket you’re in and the quickest fix.

drowsy eagle
#

Config : ~.openclaw\openclaw.json
Agent dir : ~.openclaw\agents\main\agent
Default : anthropic/claude-haiku-4-5-20251001 (from claude-haiku-4-5-20251001)
Fallbacks (0) : -
Image model : -
Image fallbacks (0): -
Aliases (0) : -
Configured models (0): all

Auth overview
Auth store : ~.openclaw\agents\main\agent\auth-profiles.json
Shell env : off
Providers w/ OAuth/tokens (0): -

  • anthropic effective=profiles:~.openclaw\agents\main\agent\auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | anthropic:default=sk-ant-a...yC_rIAAA

OAuth/token status

  • none
stone barnBOT
#

Yep — that models status output confirms you’re using an Anthropic API key (api_key=1), and OpenClaw is able to read it fine. So this isn’t a token/caching issue on the Gateway/UI side.

What causes “API usage limit reached” (Anthropic)

Almost always one of these:

  1. Billing / quota cap hit (most common with that exact wording)
  • You’ve hit a monthly spend limit / hard limit set in the Anthropic Console, or your account has no remaining credits / needs billing setup.
  1. Rate limit window hit (burst limit)
  • Too many requests/tokens in a short window (multiple chats, retries, big contexts, tool calls). This usually clears after a cooldown.

What to do

1) Check Anthropic Console limits/billing

Go to the Anthropic Console → Billing / Usage / Limits and look for:

  • a hard limit you reached (raise it), or
  • payment method / credits missing.
#

If it’s a hard quota issue, it won’t fix itself by restarting OpenClaw — you need to adjust the Anthropic side.

2) If it’s a rate limit: wait + reduce burst

  • Wait a few minutes and try again.
  • Avoid sending a bunch of messages quickly from multiple clients (webchat + discord + tui, etc.).

3) Add a fallback provider/model so you’re not blocked

So when Anthropic is limited, OpenClaw can fall back automatically:

openclaw models list --all
openclaw models fallbacks add <provider/model>
openclaw models status

(or rerun openclaw onboard to add another provider/key and set it up cleanly)