Nomira — Product One-Pager

What it does, who it's for, why it's different.

The problem

Teams now run AI coding assistants (Claude Code, Codex, Cursor) all day. Two developers on the same tools burn wildly different amounts, a single action can cost surprisingly much, and nobody can explain why. The vendor invoice shows totals; the assistants show raw usage; neither tells you where the money went or why. And most of the cost hides in cache and reasoning tokens that naive calculators ignore.

What Nomira does

Reads the logs you already have locally and answers: "why did this turn / conversation / developer cost what it did?" — with provider-accurate, cache-aware numbers.

• Per-action forensics — the most expensive turns and the reason (rebuilding

cache, long generation, big context), plus waste signals (retry loops, repeated reads, cache rebuilt instead of reused).

• The auditor — exact cost including cache-read, cache-write (5m vs 1h),

reasoning and tool tokens, across providers. On real data, ~90% of cost is cache a naive tool misses.

• Two cost worlds, labeled — subscription tools (Claude Code/Codex) show

API-equivalent value + allowance used; token-billed APIs show the actual bill.

• Team view — ship usage-only events to a self-hosted dashboard: cost by

developer, project, model, and trend — answering "why does A burn 5× B?"

• Private by design — token counts + tags only; **prompts never leave your

machine**; no proxy; no content stored (schema-enforced).

Who it's for

Mid-market, AI-native engineering teams with real AI spend, a free/paid tier, and privacy concerns — who don't have a FinOps team and won't route prompts through a third party.

Why it's different

Status

Stage-0 prototype, working today on Claude Code + Codex (Cursor keeps token data server-side; we report that honestly). Open source, MIT, Python stdlib only.

Try it

python nomira.py · python nomira.py --ship && python nomira.py --serve