The control layer for autonomous AI spend

CostWindow — your AI is spending right now.

Agents and cron jobs spend money while you sleep. CostWindow tracks every call, every CLI session, and every subscription — and refuses the next call when a process blows its budget. A hard stop, not a dashboard you have to remember to check.

Start free — 2 processes, no card · Go unlimited — £12/mo · Sign in

What CostWindow does

• It tells you when money leaves that nothing explains. Log your real provider balance and CostWindow reconciles it against every tracked call. If spend left your account that no process accounts for, you get a tracking-gap alert — plus runway: how many days your balance lasts at current burn.
• A kill switch that refuses the call. Route your calls through the CostWindow proxy and a killed process is refused before the provider is ever reached — a hard stop, not a polite request. Works on agents, cron jobs, and scripts you forgot were running.
• Budgets that enforce themselves. Set a monthly cap per process. At 100%, the next call is blocked — no manual step, no morning-after surprise. Resets on the 1st.
• One line to connect. The whole picture. Point your base URL at CostWindow — no SDK, no code changes, any language. Claude Code and Codex sessions track via a one-command hook. Flat subscriptions (Claude Pro, ChatGPT, tooling) — added once — sit in the same dashboard, so the number you see is your actual total AI spend.

How it works

1. Create a free account. No card, no checkout. Two processes tracked free, every feature included.
2. Point your base URL at CostWindow. One environment variable and every call — to Anthropic, OpenAI, Azure OpenAI, Google Gemini, AWS Bedrock, Groq, Mistral, xAI and other major providers — is tracked and enforced, any language, any tool. A Python SDK and one-command Claude Code hook are also available. Your provider API key passes straight through and is never stored.
3. Set budgets. Sleep. Cap each process monthly. Blown caps and killed processes are refused before the provider is called.

Providers & security

Supported through the one drop-in proxy: Anthropic, OpenAI, Azure OpenAI, Google Gemini, AWS Bedrock, Groq, Mistral, xAI, Together, Fireworks, OpenRouter, DeepInfra, Perplexity and any other OpenAI-compatible endpoint. Token usage is tracked for every call; cost and budgets apply to all major models, with unlisted models flagged rather than silently counted as zero. Tracking granularity is per named process, with a monthly budget enforced per process. Your provider API key passes through untouched and is never stored by CostWindow — and only token counts, model, and cost are recorded, never your prompts or responses. Full detail — the request path, exactly what is stored and what never is, and how the hard stop refuses a call: How CostWindow works.

Pricing

Every feature on every plan. Paid lifts the process cap. No seat limits.

• Free — £0. 2 processes, every feature, no card. Start free
• Monthly — £12. Unlimited processes, cancel anytime. Start Monthly
• Yearly — £99 (£8.25/mo, save £45 vs monthly). Start Yearly

Included on every plan: drop-in proxy (no code changes) · hard kill switches & auto-block budgets · untracked spend detection · provider balance runway · Python SDK (Anthropic + OpenAI) · Claude Code & Codex session hooks · subscriptions tracked alongside API spend · spike alerts & notifications.

A product by myAIcraft — part of the Window family with RadarWindow.