See where your team's
AI spend actually goes.
Lumen Pro sits between your AI tools and the model provider, records what every request was for, and rolls it into a fleet dashboard โ spend by person, project and repository.
The gap
Your AI bill says how much. Not what for.
Lumen Pro captures attribution at the point the request leaves the device, then rolls it up for the whole team.
One bill, many seats
The total arrives. The split stays hidden.
Totals without attribution
Provider dashboards show how much, never for what.
Sometimes no cost API
claude.ai Team plans expose nothing to reconcile.
Attribute spend to work
Cost by project, repository, branch and person โ the shape chargeback and budget reviews need.
- Chargeback and client billing
- Spend by credential
See usage outside policy
Approved repositories and working hours, reported as a share of spend rather than against people.
- Repository allowlists
- Working-hours signals
Ask the fleet
Plain-English answers from your own data โ only what the person asking could already open.
- โWhat did we spend?โ
- MCP tools for any agent
How it works
Four steps, on every device.
Nothing changes about how your team uses Claude Code, Cursor or a script.
Route
A small local service, lumen-core, sits between your tools and the provider.
Read
Reads model and token counts from Anthropic, OpenAI, Google and Cursor traffic.
Price
Prices each request from 126 model entries, refreshed every six hours.
Sync
Attaches project context and syncs to DataGrout every 30 seconds.
Relay mode needs nothing installed. Proxy mode trusts a certificate generated on the device โ no admin rights, and the key never leaves the machine.
Key capabilities
Built for attribution, not surveillance.
Attribution that holds up
Spend by conversation, project timeline or account โ accounts identified by a one-way fingerprint, never the credential.
Capture you control
Each device reports from usage totals up to full visibility. A fleet can narrow that, never widen it.
Policy in plain English
Write a rule the way you'd say it; Lumen Pro turns it into precise logic and checks it against the fleet.
Ask the fleet
A dashboard assistant answers โwhat did we spendโ or โwrite me a standupโ, plus MCP tools for any agent.
Fleets live in DataGrout: invite members by email, four roles, seven dashboard views.
What teams use it for
The six jobs it does.
Chargeback and client billing
Cost per project, repository and credential.
FinanceUsage outside policy
The share of spend that fell outside your rules.
Eng leadersStandups from real activity
A narrative of what happened, tickets and links out.
Eng leadersPrompt efficiency
Where prompts waste tokens, with richer capture on.
TechnicalBenchmarking a change
Run the same task twice and compare the cost.
TechnicalReconciling the provider bill
Billed versus observed, for Anthropic Console.
FinanceTwo editions
Lumen is a monitor. Lumen Pro is an accounting.
Lumen
FreeOpen source ยท runs on your machine
- One person's machine
- Live cost and token-rate gauges
- Cache savings on every call
- No account needed, data stays local
Lumen Pro
Team editionFleet ยท attached to DataGrout
- A whole team's fleet
- Spend attributed to project, repository and person
- Policy signals and rules in plain English
- A fleet dashboard in DataGrout with four roles
Pricing quoted with your team.
Privacy by design
Attribution, not observation.
Usage totals are the default. Every richer level of capture is a deliberate choice, and members can opt out.
Usage totals only
DefaultModel, tokens and cost.
Standard
Adds folder name, git remote and branch.
Full visibility
Adds prompt and reply excerpts, intent label, account fingerprint.
- A fleet can narrow what a device reports, never widen it.
- Each fleet sees only the fields it asked for.
- Credentials are never stored โ only a one-way fingerprint.
- The dashboard assistant reaches only what the asker could already see.
Product details
What ships, and where it runs.
Platforms
macOS 14+ (universal), Windows x86-64, Linux from source.
AI clients
Claude Code, OpenCode, Cursor, Claude Desktop, CLI tools and scripts.
Providers
Anthropic, OpenAI, Google and Cursor, plus any compatible endpoint.
Requirements
A DataGrout account. Proxy mode trusts the device certificate once.
Costs are estimates
Priced from published list prices, not invoices. Cursor is calibrated.
Coverage, not absolutes
Lumen sees the traffic routed through it, and reports the share it observed.
Reconciliation is Anthropic-only
claude.ai Team plans expose no cost API; there, policy share is the picture.
Lumen is open source (MIT) on GitHub. Claude, Cursor, OpenAI, Gemini and other product names are trademarks of their respective owners.
Pricing
Priced per fleet, quoted with you.
Seats, capture tier and device count all move the number, so we quote after a short conversation rather than publish a list price.
- A walkthrough of the fleet dashboard
- Capture policy set to your team's rules
- Pricing scoped to seats and capture tier
Already on the free Lumen? Nothing changes โ it stays open source and needs no account.
Talk to sales
Tell us a little about your team. We'll walk you through the fleet dashboard and quote pricing for your rollout.
