{}
tokmeter.ai
AI FinOps · Built for engineering leaders

See your AI bill. Cap it. Move on.

OpenAI, Anthropic, Azure, Vertex, Bedrock — plus the IDE seats nobody else shows you: Cursor, Copilot, Claude Code. Paste a read-only key. Real dashboard in 5 minutes.

Start free — 5 min setupSee the dashboard ↓
No credit card · No code change · Read-only setup
Works with the models your teams already use
OpenAI
Anthropic
Google
Vertex / Gemini
Azure
OpenAI
Bedrock
AWS
Mistral
Groq
xAI
Live now On roadmap OpenAI-compatible via gateway+ 40 more · full matrix →
This is the actual dashboard

Sign up and this is what you see — pre-loaded with sample data until your first sync lands.

Scroll the preview below. Every chart, anomaly, and rollup populates with your real spend as soon as you connect a provider.

Open it for real →
app.tokmeter.ai / dashboard Sample data
Operate · Last 30 days

Spend dashboard

MTD spend
$25K
vs $32K budget
Projected EoM
$25.8K
22% under
Budget burn
78%
of monthly cap
Top spender
Jordan
Data · $2,562
Spend over time
Stacked by tool, last 30 days
Anomalies & opportunities
Worth a look this week
Platform team spend ↑ 187% week-over-week
Driven by a new background-agent workflow running Claude Opus on full repo context.
Recommended: Cap Opus to ≤10% of Platform requests; route to Sonnet.
5 engineers exceeded daily $50 budget on Tuesday
All on Cursor Business with metered overages. Soft cap not enforced.
Recommended: Promote soft cap to hard cap in /guardrails.
By tool
Cursor$7,989.00
Claude Code$6,990.00
Copilot$3,496.00
Anthropic API$2,996.00
By team
By workflow
Multi-file code-gen$8,488.00
Chat Q&A$5,242.00
Inline completion$3,994.00
Agentic loops$4,494.00
Code review$1,747.00
Docs & explanations$999.00
By model
Where token spend lands across providers
Connect a provider — replace this sample with your real spend.
Start free →
Sample data shown. Your version connects to OpenAI, Anthropic, Azure, Vertex, Bedrock, Cursor, Copilot, and more.
See full provider coverage →
From a number on the dashboard → to the receipt behind it

When finance asks "what is this $25K?", click a line and see the actual request — and what it would've cost on a cheaper model.

The dashboard above is the summary. This is what's underneath every dollar: which team ran the call, which prompt, which model, and a one-line "you could've spent $$0.0008 instead of $$0.0042" so the savings conversation writes itself.

app.tokmeter.ai / requests / req_8f3a…b21c 200 OK
openai/gpt-4oteam: support-botkey: sk-ts-live-…a91fworkflow: ticket-triage2026-06-20 14:22:31 UTC
system
42 tokens
You are a support triage assistant. Classify each ticket into: billing, technical, account, other. Reply with JSON only.
user
68 tokens
My invoice charged me twice this month and I can't find a refund button. Order #A-29481.
assistant
31 tokens · finish: stop
{
  "category": "billing",
  "confidence": 0.96,
  "order_id": "A-29481"
}
cache: miss·retries: 0·guardrails: passed (3/3)·Replay request →
Cost
$0.0042
↓ $0.0061 if routed to gpt-4o-mini
Tokens in
110
Tokens out
31
Latency
412 ms
TTFT
189 ms
Eval scores
json_valid✓ 1.00
pii_redacted✓ 1.00
category_match✓ 0.96
request_id: req_8f3a…b21c
parent_trace: trc_4d92…7f
For the CFO
Every line on the bill is auditable. No more "the AI line went up, no one knows why."
For engineering
Replay any failing or expensive request without re-running prod traffic.
For the team lead
See which workflow, key, and teammate the spend belongs to — chargeback-ready.
Already using an AI gateway?
See how we compare — honestly.
vs Cloudflare AI Gatewayvs LiteLLMvs Portkeyvs Helicone
Three steps. Skip any.

Start where you are. Go as far as you want.

Step 1Start here
See the bill

Paste a read-only billing key. Spend by team, model, and project — live in 5 minutes.

No code change
Step 2
Track per request

Drop in our SDK (one import) to attribute every call to a user, agent, or CI job.

1 line of code
Step 3
Cap the spend

Point the OpenAI base URL at us. Hard dollar caps per team. Rollback is one env var.

1 env var
What you get

Three answers your CFO keeps asking.

Per-team rollup
Where is the money going?

Every dollar mapped to a team, model, and project. No more guessing from a vendor invoice.

Mechanism, not hype
Where can we actually cut cost?

We flag calls that would route safely to a mini-tier model — e.g. gpt-4o → gpt-4o-mini for classification and summary tasks. Most teams cut 20–40% by moving ~60% of traffic.

Hard $ caps
How do we stop a runaway agent?

Set a hard monthly cap per team or key. We block requests over the limit — no surprise invoices.

When you're ready for caps

Two env vars. Zero refactor.

Your existing OpenAI or Anthropic code keeps working. We sit in front, log every call, attribute the cost, and enforce your caps. Roll back by removing one line.

Read the 2-minute install →
.env
OPENAI_API_KEY=sk-ts-live-...
OPENAI_BASE_URL=https://tokmeter.ai/v1
In every plan

One platform for AI spend, governance, and quality — so you can retire the cost spreadsheet and the standalone eval tool.

Virtual keys with hard $ caps

Issue a key per team, agent, or CI job. Set a monthly ceiling. We enforce it.

Per-engineer cost attribution

See which workflow drove which dollar. End vendor-by-vendor reconciliation.

Full audit trail

Every key, budget, and credential change is logged and exportable. Ready for your security review.

Production evals

Sample real traffic against your rules. Catch regressions before users do.

Agent and MCP tracking

Cost and policy per tool, per call. Know which agent burned the budget.

Privacy controls

PII redaction, configurable retention, customer-managed keys on Enterprise.

15ms
Median added latency on non-streaming calls
50M+
Requests per tenant per month
7 days
Default trace retention. Configurable.
100%
Audit log coverage on every key, budget, and credential change

See your AI spend before your next invoice.

Free to start. 5 minutes to your first dashboard. No code change required.

Start free →Or see pricing