Step-by-Step Fix
-
Confirm the scope
- Try a different browser/device and a different network.
- If only one environment fails, the cause is usually local.
-
Refresh your session
- Sign out completely, then sign back in.
- Clear cache/cookies for the service domain.
- Try an incognito/private window with no extensions.
-
Check permissions and plan status
- Verify you’re using the correct account/workspace.
- Confirm your subscription/plan is active and assigned correctly.
-
Rule out network filtering
- Disable VPN/proxy temporarily.
- Pause ad blockers / privacy tools that may block requests.
- If you’re on a corporate network, test via hotspot.
-
Check service incidents
- Review the product status page or recent incident reports.
- If the service is degraded, wait and retry.
-
Collect evidence and escalate
- Save screenshots + exact error text + timestamps.
- Include environment details and repro steps in a support ticket.
Common Root Causes
- Expired/invalid session tokens
- Plan or permission mismatch
- Browser extensions interfering with requests
- Network blocks (VPN/proxy/firewall/DNS)
- Temporary outages
Prevention Tips
- Keep a clean browser profile for critical workflows
- Don’t stack multiple privacy extensions that rewrite requests
- Document workspace/team permissions and billing owners
- Export important settings regularly (when supported)
Why This Happens
OpenClaw agents call the Anthropic Claude API, which enforces per-minute and per-day token limits that vary by your Anthropic API tier. A free-tier Anthropic key allows approximately 40,000 tokens per minute; Tier 1 (after first payment) allows 200,000 tokens per minute; higher tiers allow progressively more. When an agent’s prompt and output tokens combined exceed the per-minute limit, Anthropic returns a 429 rate limit error. OpenClaw may retry automatically, but if the retry interval is too short or the agent sends another large request immediately, the limit is hit again. Rate limits reset on a rolling 60-second window, not at the top of each minute.
Common Mistakes to Avoid
- Using a free-tier Anthropic key for production cron jobs: Free-tier keys have very low rate limits and are unsuitable for automated workflows that run multiple times per day. Upgrade your Anthropic API account to at least Tier 1 by making a small initial payment, which significantly increases your per-minute limit.
- Running multiple high-token agents simultaneously with the same API key: Parallel agent runs share the same per-minute token budget. Two concurrent agents each using 100,000 tokens per minute will together exceed the Tier 1 limit. Stagger scheduled runs or use separate API keys for high-volume agents.
- Not adding retry logic with backoff: An agent that immediately retries after a 429 error will continue receiving 429 errors until the window resets. Configure exponential backoff (e.g., wait 10 seconds, then 20, then 40) before retrying after a rate limit hit.
- Using large context windows unnecessarily: Agents that include full file contents, long conversation histories, or large tool outputs in every API call consume tokens rapidly. Truncate context to only what the current step requires to reduce per-call token usage.
Understanding OpenClaw and Anthropic Rate Limits
OpenClaw itself may also impose its own run concurrency limits based on your subscription tier — for example, limiting the number of simultaneous agent runs to 2 on a starter plan and 10 on a pro plan. These platform limits are separate from Anthropic API rate limits. A 429 error in your run logs that mentions "rate limit" is an Anthropic API error. An error that mentions "concurrent run limit" or "quota exceeded" refers to OpenClaw’s platform limits.
To identify which limit you hit: open the failed run log and look at the full error message. Anthropic rate limit errors include the phrase "type": "rate_limit_error" in the JSON. OpenClaw platform quota errors typically appear as a plain-text message like Maximum concurrent runs reached.
Additional FAQ
Q: What are the Anthropic API rate limits that affect OpenClaw agents?
Anthropic rate limits are tiered based on your API account level. The free tier allows approximately 40,000 input tokens per minute and 8,000 output tokens per minute across all models. Tier 1 (triggered by a minimum spend of $5) allows 200,000 input tokens per minute for Sonnet and Haiku models. Tier 2 and above allow higher limits. Claude Opus has lower per-minute limits than Sonnet regardless of tier. Check console.anthropic.com > Usage Limits to see your current tier and exact limits.
Q: How long does it take for a rate limit to reset after I hit it?
Anthropic rate limits reset on a rolling 60-second window. This means that if you exhaust your per-minute token budget at 2:00:30 PM, your budget begins refilling immediately as the 60-second window rolls forward — you do not need to wait until 2:01:00 PM for a complete reset. In practice, waiting 60 seconds after hitting a rate limit is sufficient for a full budget recovery. Daily token limits (if applicable) reset at midnight UTC.
Q: Can I increase my Anthropic API rate limits without upgrading my plan?
You can request a rate limit increase directly from Anthropic by visiting console.anthropic.com > Usage Limits and clicking "Request Limit Increase." Anthropic evaluates these requests based on your account’s payment history and stated use case. There is no guarantee of approval, but accounts with consistent usage history and a clear business purpose are typically approved within 2–5 business days. Alternatively, upgrading to a higher Anthropic API tier by increasing your monthly spend automatically unlocks higher limits without a separate request.
Q: Why did my agent start hitting rate limits this week when it worked fine last week?
Rate limit issues that appear suddenly without configuration changes usually indicate that your agent’s token consumption per run increased. This can happen if: (1) the data the agent processes grew larger (e.g., a longer file, more search results), (2) you added a step that generates longer outputs, or (3) another workflow started sharing the same API key. Check your Anthropic console for a usage chart — if token consumption per minute spiked recently, the agent’s workload changed, not the rate limit.
Related Issues
- OpenClaw 401 invalid API key errors cron jobs
- OpenClaw agent not responding stuck run
- OpenClaw cron job failing find logs root cause
- OpenClaw tool calls failing timeouts not available
Additional FAQ
Q: How do usage limits actually reset — daily or rolling? Most AI platforms use either a fixed daily reset (e.g., at midnight UTC) or a rolling window (e.g., your oldest message from 3 hours ago expires and frees up a slot). Rolling windows are more common for message and request limits because they distribute server load more evenly. Check the platform's help documentation for the exact mechanism — the support page for your specific limit usually specifies the reset type and time zone.
Q: Can using a VPN bypass usage limits? No. Usage limits are tied to your account, not your IP address or location. A VPN changes your apparent location and IP, but the platform still identifies you by your authenticated account session. Attempting to bypass limits using VPNs, multiple accounts, or shared credentials violates most platforms' Terms of Service and can result in account suspension. The correct path is to upgrade your plan, wait for the limit to reset, or use the API if available.
Q: What is the difference between a soft limit and a hard block? A soft limit reduces your access gracefully — for example, automatically switching you to a lower-quality model when you reach your cap, or slowing response speed. A hard block fully stops access and shows an error message or countdown timer. Soft limits let you continue working at reduced capability; hard blocks require waiting for a reset or upgrading your plan. Most platforms implement soft limits before hard blocks to reduce user disruption.
Related Articles
- OpenClaw login / API key fix
- OpenClaw rate limits affecting runs
- OpenClaw agent not starting
- OpenClaw tool calls failing
Additional FAQ
Q: How do usage limits actually reset — daily or rolling? Most AI platforms use either a fixed daily reset (e.g., at midnight UTC) or a rolling window (e.g., your oldest message from 3 hours ago expires and frees up a slot). Rolling windows are more common for message and request limits because they distribute server load more evenly. Check the platform's help documentation for the exact mechanism — the support page for your specific limit usually specifies the reset type and time zone.
Related Articles
- OpenClaw login / API key fix
- OpenClaw rate limits affecting runs
- OpenClaw agent not starting
- OpenClaw tool calls failing
Additional FAQ
Q: How do usage limits actually reset — daily or rolling? Most AI platforms use either a fixed daily reset (e.g., at midnight UTC) or a rolling window (e.g., your oldest message from 3 hours ago expires and frees up a slot). Rolling windows are more common for message and request limits because they distribute server load more evenly. Check the platform's help documentation for the exact mechanism — the support page for your specific limit usually specifies the reset type and time zone.