What your AI agents
burn — and what
they can reach
Meter is a proxy for Claude Code that keeps the books: which developer spent how many tokens, on which repositories, and which risky actions the agent took. Visibility first, then control.
Works with Claude Code. Set up in ten minutes.
Example · sample data
Spend this week
62% above
last week
$4,218
| developer | repository | tokens | $ |
|---|---|---|---|
| a.kovalenko | payments-api | 42.1M | $612 |
| m.bondar | web-core | 38.4M | $548 |
| i.shevchenko — over budget, needs attention | infra · .env | 29.7M | $431 |
| o.melnyk | web-core | 24.3M | $389 |
| k.tkachenko | docs-site | 11.0M | $164 |
Claude Code is supported today. Cursor and Copilot are next on the roadmap.
Agents rolled out faster
than the means to govern them
The company is already paying — blind. Three things Meter makes visible.
- Spend with no owner
- The AI bill has multiplied, with no breakdown by person, team or repository. The CFO asks why, and nobody can answer.
- Access with no limits
- The agent has the same rights as the developer: it reads .env, opens private repositories, pushes to branches. No audit, no policies.
- Spend with no return
- Nobody knows who uses agents or how well. You pay, but can’t tell whether it pays off.
A transparent proxy between agent and model
Developers change only the base URL. Meter logs, attributes and enforces policies without changing how the agent behaves.
Agent
Claude Code calls Meter with the developer’s own key
Meter Gateway
Records tokens, cost, repository and action; raises security flags. Accounting runs off the request path — the agent’s response never waits on it.
LLM provider
The request passes straight through; the response reaches the developer as usual
How to connect Claude
Meter works with API keys: your agents’ requests go through it.
Anthropic API key
AvailableDevelopers change only the base URL, and requests flow through Meter.
- Actual spend from the API bill, by person and repository
- Security flags on every request
- Limits, blocking and approval of risky actions
Meter doesn’t support Claude Pro and Max subscriptions: a subscription can’t be routed through a proxy, so Meter can’t see it.
What’s inside
Spend attribution
Who, how many tokens, which repository, which model — broken down by person and team.
Security flags
Secret access, pushes to protected branches, sensitive paths — all flagged on the event.
Policies and limits
Per-team budgets, repository allow/deny lists, approval for risky actions.
Weekly reports
Spend and risk breakdowns in Slack and email — the artifact people actually read.
Self-hosted
Deploy in your own cluster. Code and prompts never leave your infrastructure.
Priced per developer, not per token
Predictable for the CFO. Start with visibility, grow into control.
Beta: everything is free for now. Checkout runs in test mode — no money is charged.
Visibility
Dashboard, attribution, security flags, warning budgets and alerts in the dashboard
Free
Get startedControl
Everything in Visibility, plus weekly reports and alerts by email and Slack, blocking budgets, access rules, guardrails and approvals
$20/developer/mo
minimum 3 seats
Platform
Everything in Control, plus SSO, roles, audit trail, self-hosted and SIEM
Custom
Coming soonComparisons
Meter beside the other tools
What each one does, and when you’re better off without us. Dated, with links to their own documentation.
- Meter vsAnthropic’s toolsConsole analytics and the Claude apps gateway: what they already give you at no extra cost, and what they lack — repositories, flags, approvals.Read the comparisonChecked 2026-10-05
- Meter vsLiteLLMThe most popular open-source LLM gateway: 100+ providers and per-key budgets. No repositories, no flags on agent actions.Read the comparisonChecked 2026-10-05
- Meter vsPortkeyAn AI gateway with observability, part of Palo Alto Networks since May 2026. Budgets on Enterprise and for select Pro customers; logs request text.Read the comparisonChecked 2026-10-05
- Meter vsHeliconeOpen-source LLM observability, in maintenance mode at Mintlify since March 2026. Request logs, no flags or approvals.Read the comparisonChecked 2026-10-05
Frequently asked questions
Does Meter see my prompts and code?
They pass through the proxy, but Meter doesn’t keep them. It keeps the call’s metadata, and the file paths and commands the agent passed to its tools.
Does Meter replace my Anthropic bill?
No. Calls go out with your team’s own key, so you pay Anthropic for tokens as before. Meter is priced per developer seat, separately from that bill.
Does Meter add latency?
Accounting runs off the request path. Under load, CI checks that Meter adds no more than 3 ms at p50 and 15 ms at p99 over a plain proxy.
Does Meter work with Claude Pro and Max subscriptions?
No: a subscription can’t be routed through a proxy, so Meter works with API keys only.
Do security flags block what the agent does?
No, a flag only marks the event. Blocking an action or requiring approval is done with policies on the Control plan.
How soon do I see data?
A developer changes only ANTHROPIC_BASE_URL and the key — the first calls show up in the dashboard within seconds.
Can I run Meter myself?
Yes, with a Helm chart in your cluster. Your data then stays in your infrastructure.