Every LLM API key, capped, routed and audited.

Day 20 of September · 6 keys · 4 Claude models · list prices from Anthropic pricing page

Month-to-date spend–

Forecast month-end–

Budget (sum of caps)–

Keys over policy–alert · throttle · block

Savings applied / mo–

API keys & policies

Drag a cap, pick an action — the policy engine re-evaluates instantly

Key / team	Model	MTD $	Forecast $	Cap $ / mo	Use	Action	State

Daily spend, stacked by key

Next action — savings you can take now

–

per month at current volume, same quality tier where marked

Tamper-evident audit log

What Meterline costs you (our revenue model)

Governed LLM spend / month

How it works

1 · ProxyOpenAI/Anthropic-compatible edge endpoint, per-key virtual tokens

2 · MeterToken counts from provider usage fields, priced by versioned rate card

3 · PolicyCap → alert / throttle / block, evaluated per request <2 ms

4 · OptimizeCache, route, batch recommendations with $ proof

5 · AuditSHA-256 hash-chained events, CSV/FOCUS export

Prompts are never stored — only token counts, model, key and latency. Rate card: docs.anthropic.com/en/docs/about-claude/pricing (fetched 2026-09-26). Seed traffic is synthetic.

Governance & compliance evidence

Framework · control	What Meterline records	Evidence	Status

Who can change what (role-based access)

Role	Edit caps	Change action	Apply savings	Export evidence
FinOps admin	all keys	all keys	yes	yes
Team lead · CX	own team only	own team only	own team only	no
Auditor	no	no	no	yes (read-only)

Switch "Acting as" in the header: controls lock instantly and every role switch is written to the audit chain. Mappings show where evidence comes from; they are not a certification.

Policy as code & live request trace

1 Authvirtual key

2 Policycap check

3 Routemodel rule

4 Forwardprovider

5 Meterusage × rate

6 Audithash-chain

Pick a key and send a 1,200-in / 300-out token request through the same policy engine.

This YAML is generated from the live table above. In production it is versioned in Git and every change is a signed audit event.

Raw data behind every number

Nothing on this page is typed by hand: each dollar is recomputed from these rows

Rate card (USD / 1M tokens)

Model	Input	Output	Cache hit

Source: docs.anthropic.com/en/docs/about-claude/pricing, fetched 2026-09-26 (rate-card version rc-2026-09-26)

Usage events, last 7 days (million tokens, from provider usage fields)

Day	Key	Model	Input MTok	Output MTok	Cost $

Seed traffic for the demo workspace is synthetic and deterministic (fixed PRNG seed), so every judge sees the same rows.