AI Agent Spending Cap

Stop your AI agents from burning money

Set a spending cap. ClawCap enforces it and picks the cheapest model that works.

3 Commands
9 Providers
<5ms Latency
Get Started

Your agents don't sleep.

But you do. And when something goes wrong at 3 AM, you need a safety net — not a laptop.

// Without ClawCap
  • Agent goes rogue at 3 AM burning your API credits
  • SSH into your server to find what's running
  • No provider has built-in spending caps
  • Away from your computer? You're stuck.
// With ClawCap
  • Kill switch from Telegram. One tap.
  • Hard cap blocks all requests at your limit
  • No more runaway API bills
  • Loop detection auto-pauses agents before damage

How it works

Install once. Protect everything from anywhere.

ClawCap
01
Install

One command: npm install -g clawcap

02
Configure

clawcap init — wizard for API keys, caps, and Telegram.

03
Start the proxy

clawcap start — proxy is live. Point your agent at localhost:5858

04
Sleep easy

ClawCap enforces caps 24/7. Check spend from your phone via Telegram /status

Layers of control

The core is simple: set a cap and it's enforced. Everything else is optional protection.

Hard Spending Caps

Set daily and monthly limits. When you hit them, requests get blocked with a clean 429. Not throttled. Blocked.

Loop Detection

Catches agents stuck in loops — repeated payloads, error floods, rapid-fire requests. Auto-pauses and alerts you.

Heartbeat Protection

Detects idle keep-alive calls and routes them to a cheap model. Blocks heartbeats during sleep hours.

Phone Kill Switch

Kill or resume agents from Telegram. Get proactive alerts at 80% and 95% of your cap. No laptop required.

9 Provider Routing

Anthropic, OpenAI, Gemini, DeepSeek, Mistral, xAI, Groq, MiniMax, Moonshot. Routes by model name.

Multi-Channel

Telegram bot, CLI commands, HTTP API — control your agents from whatever channel you prefer.

Simple pricing

Start free with core protection. Upgrade when you need alerts, wallets, and team features.

Free

$0 / forever
  • All core features
  • Unlimited requests
  • Daily + monthly caps
  • Loop + heartbeat detection
  • CLI control
  • Self-hosted
  • No Telegram alerts
  • No prepaid wallet
Start Free

Team

$29 / seat / month
  • Everything in Pro
  • Team spend dashboard
  • Per-agent cost allocation
  • SSO integration
  • Webhook notifications
  • Custom cap policies
  • Dedicated support
  • SLA guarantee
Contact Us

FAQ

Everything you need to know.

Yes. ClawCap is a transparent proxy. Any tool that can point to a custom API base URL works — OpenClaw, LangChain, AutoGPT, custom scripts, anything that calls Anthropic or OpenAI-compatible APIs.

Under 5ms. ClawCap runs on localhost and checks are in-memory / SQLite. The bottleneck is always the upstream LLM API, not the proxy.

ClawCap reads token counts from API responses and multiplies by the model's known pricing. Costs are stored in a local SQLite database. No data is sent anywhere.

Keys are stored in your local .env file and never leave your machine. ClawCap runs entirely on your infrastructure. No telemetry, no external calls, no phone-home.

ClawCap returns a 429 status with a clear JSON error message explaining the cap was reached. Your agent gets a clean signal to stop, not a mysterious failure.

No. Telegram is optional. You can control everything via the CLI with clawcap status, clawcap kill, and clawcap resume.

Yes. ClawCap includes a Dockerfile for containerized deployment. Build with docker build -t clawcap . and run with docker run -p 5858:5858 --env-file .env clawcap.

Sign up for a plan, install the CLI with npm install -g clawcap, then run clawcap init for interactive setup. You'll be protected in under 2 minutes.

Ready to take control?

Set your cap. Point your agent at ClawCap. From that point on, your agents can't burn more than you allow.

Get Started