Phase attribution for AI coding
Bill-y.
AI coding spend, measured by workflow — not just volume.
See whether AI is used well, not just heavily.
Bill-y splits an engineering team's AI token spend across the phases of building software — so leaders can see where the money goes, and whether it is working.
The idea
Engineering teams now spend hundreds of dollars per developer, every month, on AI. But a token count tells you how much you spent — never whether it was spent well.
Bill-y attributes every token to a phase of work: understand, design, build, test, review. Cost stops being a number you pay and becomes a signal you can act on.
What it does
Phase attribution
Splits spend across understand, design, build, test and review — the real shape of how the work happens.
Cost per outcome
Measures cost per outcome instead of raw usage, so spend is paired with what actually shipped.
Purpose-built for Claude Code
Deep capture of real coding sessions, not a shallow API meter bolted on after the fact.
Runs locally
An efficiency coach, not a surveillance dashboard. Raw session text never leaves the machine.
Distribution analysis
Catches inefficiency the totals hide — blind generation, missing design, rework loops.
Editions
Bill-y for Enterprise
On-prem · ZDR-ready
Fleet-wide phase analytics, healthy-band benchmarks and governance — built for organizations that can't send telemetry off their own infrastructure.
Request early access →Bill-y for Developers
Free · coming soon
A local phase dashboard for your own Claude Code usage. See where your tokens go, on your machine, with nothing to sign.
Notify me →How it works
Observe
Capture live usage from real coding sessions, on your own infrastructure.
Measure
Classify each turn into a workflow phase with an on-device model.
Attribute
Visualize where spend goes and pair it with the outcomes downstream.
Request early access
Pilots are opening for teams running Claude Code at scale. Tell us a little and we'll reach out.