Argus vs a cron job and a spreadsheet

Polling balance endpoints yourself is a weekend project. Here's the honest list of what it takes and what you end up maintaining.

For two or three providers with a balance endpoint, a cron job writing to a spreadsheet works and costs nothing. Past that you're maintaining a connector per provider, each with its own auth header and response shape, plus burn-rate math, alert de-duplication and per-call capture for the providers that have no billing API. That's what Argus is.

The weekend version

# crontab: hourly balance snapshot to a CSV
0 * * * * ~/bin/snapshot.sh >> ~/api-balances.csv

# snapshot.sh
ts=$(date -u +%FT%TZ)
or=$(curl -sf https://openrouter.ai/api/v1/credits -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  | jq '.data.total_credits - .data.total_usage')
fc=$(curl -sf https://api.firecrawl.dev/v2/team/credit-usage -H "Authorization: Bearer $FIRECRAWL_API_KEY" \
  | jq '.data.remainingCredits')
echo "$ts,openrouter,$or"
echo "$ts,firecrawl,$fc"

Chart the CSV and you can see a balance fall. That's genuinely useful, and if it's all you need, keep it.

What it grows into

You needDIYArgus
A connector per providerYours to write and keep working28 maintained connectors
Auth quirksX-KEY, APIKEY, apikey, ?api=, x-api-version…Handled
Burn rateDiff readings, ignore top-upsBuilt in
Providers with no billing APICapture and price each callSDK and rate catalog
AlertsFire once, re-arm on clearEmail, with thresholds per rule
Agent accessWrite an MCP serverHosted MCP server
CostYour timeFree to $1,000 tracked, then $39 and $149 a month

Where DIY is better

  • Free, and nothing leaves your machine. No third party holds your provider keys.
  • Exactly the fields you care about, in the tool you already use.
  • No vendor to outlive.

Where Argus is better

  • Every quirk is already handled: Hunter's available is an allowance, Tavily reports usage, PDL hides the balance in headers. The integration pages list them all.
  • Per-call cost for providers with no billing API, priced from a maintained rate catalog.
  • Runway, spend spikes and rate-limit headroom as alerts, not columns you have to read.
  • Your coding agent can query it over MCP.

Questions

Is a cron job enough to track API balances?
For two or three providers with balance endpoints, yes. It stops being enough when you need per-call cost, providers without a balance API, burn rates that ignore top-ups, alerts that don't repeat every hour, or a view your coding agent can query.

Related

Let your coding agent watch the meter.

Connect a provider key, point Claude Code or Cursor at the Argus MCP server, and ask what the last run cost.