Skip to main content

Let agents run through the whole job.
Rate limits stop getting in the way.

OpenClaw, Hermes, Pi, Conductor, and other agents can stall during long tasks when they encounter rate limits. OpenLimits distributes requests across a pool of accounts so that most of these interruptions are avoided. A single key provides access to GLM, MiniMax, DeepSeek, Claude, Codex, GPT-5.6, and Fable (depending on the plan).

tokens last 24h
requests last 24h
cache hit rate
avg response time

Agents use different models for different steps.

Quick loops are well suited to the faster models, while more demanding steps benefit from stronger models. OpenLimits provides a single key and dashboard while allowing you to select the appropriate model group for each workload.

Open, agent, and premium models. One key.

GLM, MiniMax, and DeepSeek are suitable for fast loops. Claude or Codex can be used for steps that require deeper reasoning. The same endpoint, dashboard, and history are used regardless of the model group selected.

Short passes from €10/day, monthly from €120

Effort Levels

The effort parameter (low, medium, high) is available by default. It can be set on a per-request basis to balance speed against output quality.

Your Own Dashboard

Analytics are updated in real time. The dashboard includes a live request feed, per-model breakdowns, and detailed token counts for input, output, and cache operations.

Works With Everything

A single key is compatible with the tools you already use, including OpenClaw, Hermes, Pi, Conductor, OpenCode, Claude Code CLI, Codex Desktop, Cursor, direct API calls, and other clients that support a custom base URL.

No More 4-Hour Cooldowns

Users on individual accounts are sometimes locked out for hours after a short burst of activity. OpenLimits routes requests across multiple accounts so that workflows continue without interruption.

How access stays open

No custom model or special tricks are involved. Access is provided through pooled enterprise accounts and intelligent routing.

Bulk enterprise capacity

Enterprise access is purchased directly from the providers. No borrowed or scraped credentials are used. All requests are sent to the real APIs using legitimately provisioned accounts.

Multiple providers, one key

Requests are distributed across multiple provider accounts. Because no single account bears the full load, rate limits and cooldowns that affect individual accounts are avoided.

Smart routing

Traffic is distributed across eligible accounts. When one account becomes slow or rate-limited, requests are automatically directed to the next suitable account.

Full models, zero added layers

GLM, MiniMax, DeepSeek, Claude, Codex, GPT-5.6, and Fable are available with no additional prompt modifications or refusal layers introduced by OpenLimits. Data is forwarded as provided.

Fast loops on lower-cost models and more demanding passes on stronger models are both supported through the same API key.

1,800 requests per minute.
No concurrency limit.

The guard is 30 requests per second (1,800 per minute). No limit on how many can run at once.

ProviderRequests / minConcurrencyCooldowns
OpenLimits1,800UnlimitedNone
z.ai60Token-limited429 errors
Anthropic API Tier 150Token-limited429 errors
Anthropic API Tier 44,000Token-limited429 errors
Claude Pro / MaxN/A~5 messages5h & 7d lockouts
OpenAI API~3,500Token-limited429 errors
xAI / GrokVariesToken-limited429 errors

Anthropic Tier 4 still has per-model caps. Claude Pro and Max cut you off after a handful of messages. OpenLimits has no per-minute token caps, no concurrency cap, and no cooldown windows.

Three steps. About thirty seconds.

See the full instructions if you want the details.

Sign up and get your key

Create an account. The key appears right away. It unlocks the model groups on the plan you pick.

~10 seconds

Set one environment variable

Tell your tool the new endpoint. One line in the shell profile is enough.

~10 seconds

Code like normal

Use OpenClaw, Hermes, Pi, Conductor, OpenCode, Claude Code CLI, Codex Desktop, Cursor, or any other compatible client. Requests are routed through your key automatically.

~10 seconds
Flexible plans for agents
Start for as low as €10
No rate limits on the pool. Cancel when you want.
Day Pass • Week Pass • Max • Ultimate

Plans that match how the work actually goes.

Day PassLimited Time Offer
€10/ 24h

Good for trying it out or one long day of work. Access ends after 24 hours.

Start for €10
Includes
  • Access for 24 hours
  • GLM, MiniMax, DeepSeek, Claude, and Codex groups
  • Streaming, tools, images, extended thinking
  • Dashboard and analytics
Week Pass
€45/ week

A full week for a focused stretch of work.

Start Week Pass
Includes
  • 7 days of access
  • All model groups
  • Analytics and live feed
  • Cancel anytime
Save with yearly billingComing soon

Built for people who ship code

Real routing and real numbers you can watch.

Full Streaming

SSE streaming that goes straight through. No extra wrapper.

Token Analytics

Input, output, cache reads, and cache writes tracked per request.

Model Breakdown

See usage split by model plus daily trends and rough savings estimates.

Live Feed

Watch requests as they come in. Filter by model.

No cooldown interruptions

Long runs without the 5-hour or 7-day cutoffs.

Native Agent Model Access

GLM, MiniMax, DeepSeek, Claude, Codex, GPT-5.6, and Fable through one key.

Common questions

Is there a catch?

No hidden fees. The plans renew on their schedule. You can cancel from the billing portal.

Is this built for agents?

Yes. It is aimed at OpenClaw, Hermes, Pi, Conductor, OpenCode, Cursor, Claude Code CLI, and Codex Desktop.

What models do I get?

Day and Week: GLM, MiniMax, DeepSeek, Sonnet, Haiku, Codex mini. Max adds Opus, the full GPT and Codex set, and GPT-5.6. Ultimate includes unlimited Fable 5.

Can this get me banned?

No. We use enterprise accounts we bought directly. Your own accounts stay out of it.

Does it work with Claude Code CLI?

Yes. One environment variable. Extended thinking, streaming, and the rest carry over.

What about rate limits?

No 5-hour or 7-day caps. A global 30 requests per second guard sits well above normal use.

Is setup actually 30 seconds?

Yes. A key and a single environment variable. The process has been timed on multiple occasions.

Can I track my usage?

Dashboard with analytics, token counts, and a live feed.

Stop hitting limits. Start shipping.

From €10/day or €120 a month. GLM, MiniMax, DeepSeek, Claude, Codex, GPT-5.6, and Fable on the higher plans. Your own dashboard.

Choose your plan
OpenLimits — Model Access for Coding Agents