Track API spend across providers in one place

What each provider gives you

This is the part nobody tells you up front. Click through for the endpoint, auth and response fields.

Build the ledger

1. 1
   List every metered API you call

   Grep for hostnames (api.openai.com, openrouter.ai, api.firecrawl.dev) and for *_API_KEY in env files. Include crons, backfill scripts and eval harnesses. They're where forgotten spend lives.

2. 2
   Pick a source per provider

   • Spend series (OpenAI, Anthropic): daily cost buckets with an admin key. Authoritative, but a day behind.
   • Balance or quota (OpenRouter, DeepSeek, Firecrawl, Hunter, Apollo…): poll it hourly and treat each drop as spend.
   • Cost on the response (OpenRouter, Perplexity, Exa): sum it per call. Exact and immediate.
   • Nothing (Gemini, Parallel, Google Places, Postmark…): count calls and units yourself and price them.
3. 3
   Normalize to one daily ledger

   One row per provider per UTC day. Keep the native unit next to USD, because Firecrawl credits, Hunter searches and SecurityTrails queries aren't dollars.

   day         provider    source             spend_usd   balance   unit
   2026-10-16  openai      costs api          31.18       —         usd
   2026-10-16  anthropic   cost report        18.35       —         usd
   2026-10-16  openrouter  balance delta       7.09       96.72     usd
   2026-10-16  firecrawl   balance delta       1.21       1742      credits
   2026-10-16  perplexity  sum of usage.cost   0.84       —         usd

4. 4
   Derive the numbers you act on

   Month-to-date spend, a linear month-end projection, burn per day over the last seven days, and days to zero for anything prepaid. See estimating runway for the burn-rate details that trip people up.

Where it goes wrong

• Top-ups read as negative spend if you diff a balance naively. Count only decreases.
• Daily buckets lag. Today's OpenAI and Anthropic numbers are partial until the day closes in UTC.
• Billing on a match: People Data Labs, LeadMagic and Prospeo don't charge for misses, so call counts overstate spend.
• Different exhaustion errors (402, 429, 400, even 401) mean a dry balance can look like a bad key. The full list.

Or let Argus keep it

Argus is this ledger, already built: it polls each provider hourly, accepts per-call events from the SDK, and serves the result to you on one dashboard and to your coding agent over MCP. If you'd rather run it yourself, here's what the DIY version takes.

Questions

Is there one API that returns spend for every provider?

No. Each provider meters differently: some expose a balance, two expose a spend series, some put the cost on every response, and some expose nothing. A cross-provider view means combining all four kinds of source.

How do I track providers that have no billing API?

Capture each call at your app's egress and price it: tokens times the model rate for LLMs, or the endpoint's per-request or per-record price for everything else.

Related

OpenAI API usageAnthropic API usageOpenRouter API usageFirecrawl API usageApollo.io API usage