AI coding cost management
Your team's AI coding spend, before the invoice.
Claude Code, Cursor, Codex, Gemini — every developer, every tool, one number. Set a budget, get pinged the moment a runaway agent burns $80, and hand finance a clean export. Local-first: no proxy, your code never leaves the machine.
Free CLI and complimentary dashboard sync for the current month during early access. Estimates audited against official vendor pricing — run lmspend --explain to check the math.
$ lmspend team AI Coding Spend — June 2026 team total: $3,418 budget $3,000 over by $418 — alerted Jun 22 by member by tool alex@ $1,204 claude-code $2,140 priya@ $986 cursor $902 sam@ $772 codex $376 jordan@ $456 most expensive day: Jun 24 — $291 (alex, long refactor) estimates at API list prices · export: lmspend export --csv $ ▌
The problem
The person who owns the AI budget is the one flying blindest.
In 2026 a developer runs 2–3 coding agents at once, each billed separately. Nobody adds it up — until the invoice does.
Surprise invoices
One long agentic session can burn $40–$80 in an afternoon. You find out at month-end, when it's too late to do anything about it.
Every vendor shows a slice
Cursor shows Cursor. Anthropic shows Anthropic. Nobody shows the total across tools — and the total is what your card gets charged.
Teams can't see per-seat
Five developers, three tools, usage-based overages. Who's driving the bill? Which project? The budget owner has the least visibility of anyone.
For teams
The layer the free trackers don't have.
Plenty of tools show one developer their own numbers. LMSpend is built for the person who signs off on the spend.
Budgets before the invoice
Set a monthly cap. Get an email and a Slack ping the moment the team crosses it — not four weeks later.
Runaway-session alerts
Any day that runs 3× your average triggers an alert. Catch the $80 debug loop the day it happens.
Per-member, per-tool roll-up
One dashboard: who spent what, on which tool, on which project. Invite by link — members join on their own free accounts.
Finance-ready export
One CSV your finance person will actually accept — by tool, model, and day, project names hashed.
Audited estimates
Costs computed from real token counts × official vendor pricing, including cache rates. Show the math with --explain.
Local-first, no proxy
We read logs your tools already write. We never sit in the request path, so we never see your code or prompts.
How it works
Ten seconds to your real number.
1 · Run it
$ npx lmspendNo config, no account. It reads the usage logs your tools already write locally.
2 · See the whole picture
$ lmspend reportTotal spend by tool, model, project, and day, across every agent you run. Deltas vs. last month.
3 · Put it on the team dashboard
$ lmspend syncOpt-in: send aggregates to your workspace for history, budgets, alerts, and the team roll-up.
Privacy
Local-first, or it doesn't ship.
- The CLI is open source (MIT) and makes zero network calls by default.
- Syncing is opt-in and sends aggregates only — token counts, costs, hashed project names. Never code. Never prompts. Never file paths.
- We're not a gateway. Your traffic never routes through us — there's nothing for us to leak.
Pricing
The report is free. Teams pay for foresight.
Complimentary Free access is available now. Solo and Team plans open when billing launches.
- Local reports across every tool
- Share cards
- JSON output & --explain
- History & trends
- Budgets & alerts
- Everything in CLI
- History & trends
- Monthly email report
- Budget & runaway alerts
- Team roll-ups
- Everything in Solo
- Team workspace & per-member roll-up
- Slack budget alerts
- Per-project attribution
- CSV / expense export
FAQ
Fair questions.
How is this different from the free usage trackers?
Free CLIs show one developer their own numbers — and they're great at that. LMSpend adds the team layer built for whoever owns the budget: caps, alerts before the invoice, per-member and per-project roll-ups, Slack, and finance-ready exports. The individual report stays free.
Is this an LLM proxy or gateway?
No. We never touch your traffic. The CLI reads usage logs your tools already write to disk, after the fact.
How accurate are the numbers?
They're estimates from your actual token counts × published model pricing (including cache read/write rates), verified against the official Anthropic, OpenAI, and Google pricing pages. Run lmspend --explain to see the exact math per model.
Which tools are supported?
Log parsing for Claude Code, Codex CLI, Cline, and Roo Code. Cursor via its official Team Admin API (invoice-exact). Flat-rate subscriptions (Copilot, Windsurf) via one line of config. Anything else — Gemini CLI, Aider, gateways — via lmspend import.
Do my teammates each need to pay?
No. On the Team plan the owner pays; teammates join by invite link with their own free accounts and sync their aggregates. One buyer, up to five seats.
Find out what your team actually spent last month.
Start free on your own machine. Team billing will be available at launch.