Cap your AI spend
before it caps you.
Live today: a hand-checked price table for every major LLM, a JSON API, and an MCP server for Claude. On the way: the hosted proxy that watches your bill in real time and pulls the plug at your limit.
- OpenAI $18.20
- Anthropic $5.91
- DeepSeek $0.60
- OpenAI
- Anthropic
- Google Gemini
- DeepSeek
- Mistral v0.2
- xAI Grok v0.2
Want to see it act in real time? Try the interactive demo →
Your AI bill is one bug, one wrong loop, one curious afternoon away from being painful.
You're using Cursor, ChatGPT API, Claude, an agent — anything that calls a model. You added your card so it would "just work." Then one quiet Tuesday you check the dashboard and you've spent $320. The platforms have a 5-day-delay invoice. By the time you notice, it's done.
The provider dashboards lag by hours, sometimes days.
OpenAI's usage page is delayed. Anthropic shows yesterday. By the time you see the spike, you've already paid for it.
"Spend limits" only email you. They don't stop you.
OpenAI's monthly budget sends a warning at 75%, 90%, 100% — and keeps charging. Anthropic does the same. There is no hard cap.
You shouldn't need to write code to control your own card.
You're not building a SaaS. You're using AI tools. You want a dial that says "never more than $X this month". That's it.
A wall between your tools and your wallet. Set up in two minutes.
Connect your provider key
Paste your OpenAI, Anthropic, Google or DeepSeek API key. We store it encrypted. You can rotate or revoke any time.
Set your monthly cap in dollars
One number: "never more than $40 this month". Optional alerts at 50%, 80%. Resets the 1st of every month.
Swap your key in your tools
Paste the UsageWall key into Cursor, your ChatGPT client, your agent. Same API shape — every call routes through us first.
We pull the plug at your limit
Real-time spend tracker. The moment you'd cross your cap, the request gets a clean "budget reached" response. Your card is safe.
Built for everyone paying an AI bill in 2026.
Live pricing, hand-checked
OpenAI, Anthropic, Google, DeepSeek, Mistral, xAI, Meta. Cross-referenced monthly with each provider's own pricing page. Anything that drifts >20% from seed is flagged.
Free public API
One GET /api/pricing away. Build your own dashboards, cost calculators, billing alerts. CDN-cached, no auth, no rate limit on read.
MCP server for Claude
Install once. Ask Claude "how much does Sonnet 4.6 cost vs GPT-5?" and it answers from our live table. Works in any MCP-compatible client.
Hard cap (in development)
OpenAI's "spend limit" sends an email and keeps charging. The hosted proxy we're building will stop the request mid-flight. Local proxy already works for single-user OpenAI — see GitHub.
Open about what's not built
We won't claim a feature is live until you can audit it running. The schema and proxy code are public. The hosted product is on the waitlist. No fake metrics.
Free today. Paid tiers when the proxy ships.
The pricing table, JSON API, and MCP server are free forever — they're our public-good layer. The hosted hard-cap proxy is the part we'll charge for. Prices below are the planned tiers; nothing is live until you can subscribe and see it work.
✓ Live now
- Live pricing for every major LLM
- Public JSON API (
/api/pricing) - MCP server for Claude + clients
- Local single-user proxy (open source)
⌛ Waitlist
- Up to 5 hosted API keys
- Hard cap up to $500 / month in spend
- Email alerts at 50% / 80% / 100%
- 90-day history + CSV export
- Per-model breakdown
⌛ Waitlist
- Unlimited hosted API keys
- Hard cap, no monthly spend ceiling
- Slack / Discord webhook alerts
- 1-year history
- Up to 5 team members
The Free tier stays free even after Plus/Pro launch. Provider charges go straight to OpenAI / Anthropic — we never mark them up.
If a tool takes an API key, it takes ours.
Same key format. Same endpoints. Same response shape. Swap one env var and your tool keeps working — only now with a hard dollar cap.
Code editors
- Cursor
- Continue.dev
- Cline / Roo Code
- Aider
- Zed
- VS Code Copilot Chat
Chat clients
- TypingMind
- BoltAI
- LibreChat
- BetterGPT
- Any client that takes
OPENAI_API_KEY
Agents & scripts
- LangChain / LangGraph
- LlamaIndex
- Vercel AI SDK
- OpenAI Agents SDK
- Your
curlor Python script
Don't see your tool? If it accepts OPENAI_API_KEY as an
environment variable or a settings field, it works.
Answered honestly.
What works today vs what's coming?
Today (free, live): the live pricing table for every major LLM, a public JSON API, and an MCP server you can install in Claude or any MCP client. Coming: the actual hard-cap proxy (encrypted keys, real-time spend metering, alerts). The waitlist is for the proxy.
Will my API key be safe when the proxy ships?
That's the plan. The schema is already public in our repo: v2-product-schema.sql. Keys will be encrypted at rest with AES-256-GCM and the encryption key will live separately in env. We haven't shipped the proxy yet, so we won't claim it works until you can audit it running.
Will the proxy see what I'm asking the LLM?
No. By design we'll only count request and response sizes in tokens to do the math. Prompts and responses pass through as bytes — we don't buffer or read them. The local proxy already works this way; the hosted one will follow the same rule.
Can I try the proxy locally now?
Yes — clone the repo and run node proxy/server.js.
See proxy/TRY.md.
It's single-user, OpenAI-only, stores spend in ~/.usagewall/spend.json.
Streaming token accounting is not yet supported in the local
version; the hosted one is on the roadmap.
Why isn't there a hard cap from the providers themselves?
OpenAI's monthly budget sends an email warning and keeps charging. Anthropic does the same. Both confirmed there is no API-side hard cap as of May 2026. That gap is what we're building for. Until then, the safest workaround is your own proxy plus our pricing table — see /models.
How current is the pricing table?
Hand-checked monthly against the providers' own pricing pages,
with an OpenRouter live overlay. We mark prices as
estimated when a model is announced but pricing isn't
public yet, and flag any feed value that drifts >20% from the
seed. Full source: /api/pricing.
Is UsageWall a real company?
It's a product by Baku, a tiny dev shop in Santiago, Chile (Pablo Faúndez). It's not a funded startup — that means moves are slower but also that we have no pressure to fake metrics or scale before we're ready.
How do you handle my waitlist data?
We store your email, an HMAC hash of your IP (never the raw IP), your truncated User-Agent, and what you wrote in the optional "what are you building" field. That's it. Full detail: /privacy. Unsubscribe link in every email.
Tell us what you'd want capped.
The pricing layer is free and already live. The hosted hard-cap proxy is the part we're building. Join the waitlist and we'll email you once when it's shippable — no marketing spam in between. We'll ask one question: what tool or agent is burning the most on your card?
No spam. One email when we launch.