THE DOCS

Hi. If you're reading this, you probably noticed your AI bill is going up and you can't quite explain why. Maybe a teammate ran an agent in a loop overnight. Maybe Claude Opus is being called when Haiku would have done fine. Maybe a single client project ate more than the others and you can't prove it.

Burn Limit is built for exactly that. We're a small product, not a platform. Read this whole page once and you'll know everything we do — and everything we don't.

We were spending hundreds of dollars a month on Claude and OpenAI as solo developers, and the only way to know how was to log into three different dashboards and math it ourselves. None of them told us "you spent $4.20 on this side project today," or "Sonnet would have done that for 60% less," or "this client owes you $312."

The big observability tools (Datadog, Langfuse, Helicone) are built for ops teams at Series B+ companies. Brilliant tools, but overkill if you're one person trying to figure out whether the AI side of your work is profitable.

Burn Limit is the small, opinionated, slightly retro tool we wanted for ourselves.

You can be up and running in about 90 seconds.

We give you three on-ramps because every developer's setup is different.

Pricing tables are kept up to date in-app under the Models page. Today we cover:

• Anthropic — Claude Opus 4.7, Sonnet 4.6, Haiku 4.5, and older 3.x families
• OpenAI — GPT-4o, GPT-4o-mini, o1, o3, and the rest of the GPT family
• Google — Gemini 2.5 Pro, Gemini 2.0 Flash
• Kimi (Moonshot)
• Cursor — via extension, for tracking Cursor's built-in usage

Free plan is capped to one provider at a time. Dev plan and above is unlimited.

Here's what each tier actually gets you. No tier exists to push you to the next one.

Billed monthly through Lemon Squeezy. Cancel anytime — no annual lock-in.

If you're shipping AI features to clients and trying to make a margin on it, this is the section to read carefully.

1. Create a Project per client engagement. Tag it with the client name and a budget.
2. Sessions tagged to that project roll up automatically. You'll see exactly how much AI cost each project consumed.
3. When the engagement is done, generate a Pass-through Invoice. Set a markup percentage (we default to 0% — you decide). We produce an invoice with line items, totals, and the markup made transparent.
4. Share the invoice with a public link, or export it as PDF / CSV for your accounting software.

For ongoing clients, give them a Client Portal link with your branding. They see live spend, not stale monthly PDFs. It builds trust and saves you a status meeting.

Team plan is for the lead who got blindsided by a $4,000 OpenAI bill in March and doesn't want a repeat.

• Team dashboard — every member's spend in one view, sortable by cost or model.
• Team budget alerts — set a monthly cap. We email you at 80% and again at 100%. The check runs daily at 9am UTC.
• Spike alerts — if any member burns 3× their normal hourly rate (and at least $0.50), we email within the hour. This is the runaway agent guard.
• Efficiency Advisor — tells you, per developer, what they'd save by downgrading specific calls to a cheaper model. Concrete numbers, not general advice.
• Monthly digest email — on the 1st of each month, the team owner gets a one-page summary: total, vs last month, top spenders, top models, budget utilization.

Add team members by email under /team. The invitee gets an email telling them the team owner can now see their AI usage data. They can leave the team from their settings any time — we're not in the business of surveillance.

The truthful version, not the legal version (that one's on the Privacy page).

• We store your token counts, model names, timestamps, and costs. That's what powers your dashboard.
• We do not store your prompts or completions. If you're using the local proxy, content stays on your machine — only the metadata leaves.
• We do not sell or share your data. Ever. There's no "industry benchmarks" product on our roadmap that requires aggregating across customers.
• Your API tokens (OpenAI / Anthropic admin keys) are used at request time and not persisted after the import job completes.
• Account deletion is real. Click "Delete Account" in Settings and we cascade-delete everything: usage records, projects, invoices, branding, audit logs, and your auth identity. There's no soft-delete or 30-day grace period.

Some honesty about gaps. We'd rather you know now than discover them on day three.

• No hard spend caps that block API calls. We alert you, but we don't sit in front of your providers and refuse requests. Adding a hard shutoff to the local proxy is on the list.
• No Slack or webhook integrations. Alerts are email-only today.
• No SAML / SSO. Login is email + password or Google. If you need SAML, write to us — we'll prioritize it if there's real demand.
• No mobile app. The web app is mobile-responsive, but there's no iOS / Android wrapper.
• No on-prem. We're a hosted SaaS only.

So you don't hit them by accident:

• API (everything): 100 requests per minute per IP
• File upload: 10 per hour per user, 10 MB max, .jsonl only
• Admin import: 5 per hour per user, 92-day window per call
• Extension sync: 30 per hour per user
• Proxy log: 1,000 per minute per user

If your real workflow needs more, email us — we can lift limits per-account.