AI cost tracking / 2026-05-09 / 7 min read
AI Cost Tracking for Indie Builders: What to Measure Before Routing
A practical framework for tracking LLM spend by workflow, model, lane, tokens, and budget risk before enabling automatic routing.
Thesis
Routing should not be the first step. The first step is knowing which workflow creates spend, which model handled it, and whether the budget risk is visible before the invoice arrives.
Audience: Indie builders spending $50-$500 per month on agents, automation, extraction, classification, or coding assistants.
Start with workflow-level cost, not provider totals
Provider dashboards are useful for billing, but they usually collapse everything into account-level spend. That hides the product decision that matters: which workflow is consuming the budget.
For an indie builder, the core unit should be a workflow such as classification, extraction, coding assistance, realtime UX, or hard reasoning. Each workflow should have a model, lane, token count, estimated cost, baseline cost, and status.
Track enough fields to explain the invoice
A useful AI cost tracker should capture provider, model, input tokens, output tokens, workflow, optimization goal, route name, call count, latency, estimated cost, and baseline cost.
This is enough to answer practical questions: which workflows can move to a cheap lane, which ones need premium reasoning, and which ones are growing faster than the budget.
Budget alerts come before automatic routing
Automatic routing sounds attractive, but it is risky if you cannot already trust the cost report. A bad router can save pennies while breaking product quality.
The safer sequence is observe-first: log events, generate a weekly savings report, set a budget threshold, and only then test cheap, fast, and premium routing lanes on low-risk workflows.
What AgentCosts measures first
AgentCosts Router starts with persistent usage events, Cloudflare D1 storage, provider readiness, and weekly savings summaries. This keeps the product cost-first instead of becoming a generic gateway.
The goal is not to replace every provider dashboard. The goal is to show where your AI budget goes by workflow before the invoice hurts.
Decision checklist
- Assign every AI call to a workflow.
- Record provider, model, input tokens, and output tokens.
- Estimate cost and baseline cost for each event.
- Group events into weekly savings reports.
- Set a monthly budget threshold and alert before risk becomes spend.
- Only test routing after the report explains the waste.
Related AgentCosts workflows
FAQ
What is AI cost tracking?
AI cost tracking is the practice of recording model usage, token volume, workflow, and estimated spend so builders can understand which product actions create AI costs.
Should I enable model routing before cost tracking?
No. Cost tracking should come first because routing decisions need evidence about which workflows are expensive, which ones are quality-sensitive, and where budget risk is growing.
What should indie builders track first?
Start with workflow, provider, model, input tokens, output tokens, estimated cost, baseline cost, and monthly budget usage.