AI coding spend, reconciled to shipped code

Find out what your AI coding spend actually shipped.

TokenSpend reconciles Claude Code usage against GitHub to show what became merged code, what is still in flight, and what never tied back to shipped work.

See live demo

No engineer install to start·Metadata only — no source code·GitHub App, not a PAT

March 2026 · Acme Engineering

$48.2K Claude spend

Mixed confidence

$28.4K became shipped code. $9.8K did not tie to any merged PR.

Shipped
HIGH

$28.4K

59% · Attributed to merged PRs

In flight
MED

$7.1K

15% · Awaiting PR or review

Unmatched
LOW

$9.8K

20% · No merged PR after 30 days

Other
LOW

$2.9K

6% · Other repos or non-PR work

Sample data. Quick Start produces MEDIUM-confidence correlation; the plugin upgrades selected teams to HIGH-confidence attribution.

How it works

Five minutes to your first spend review.

Admin-owned integrations first. Invite engineers to add the plugin only where you need high-confidence PR attribution.

01

Connect Claude + GitHub

Pull real spend and PR activity into one place. No per-engineer install required to begin.

02

See what shipped

Every session resolves to a spend state — merged, in-flight, unmatched, non-PR, or other-repo.

03

Add precision selectively

The local plugin posts token counts and git refs only, upgrading correlation to attribution.

Team-first by default

A spend-review tool, not an engineer leaderboard.

The default view is teams, repos, sessions, and PR outcomes. Per-engineer detail is admin opt-in — and the agent never flags an individual.

Waste review, not cost mapping

Most tools show what you spent. TokenSpend shows what shipped — and what still needs review.

No install to start

Quick Start works off Claude and GitHub data before any plugin rollout.

Confidence on every number

HIGH = branch + SHA match · MEDIUM = correlation · LOW = ambiguous or unmatched.

No source code uploaded

Precision mode sends token counts, timestamps, and git refs — never your code.

Start with one repo and one pilot team.

Connect real usage, sync real PRs, then decide where precision attribution matters.