Claude Code cost tracker
Pulse is the local-first cost tracker built specifically for Claude Code. It reads your ~/.claude/projects/*.jsonl logs directly, computes cost per request at Anthropic's published rates, and tells you whether your Claude Max plan is paying off.
Last updated · 2026-05-15
Most "Claude cost tools" are export-uploaders or API-key dashboards. Pulse is different in three ways that matter when you're tracking heavy Claude Code usage:
- Local parser, no upload. Token counts are extracted from the same log files Claude Code writes to disk. Nothing leaves your machine. You don't paste an Admin API key, you don't share a workspace, you don't grant access to an OAuth scope.
- Cache TTL pricing split correctly. Anthropic bills 5-minute cache writes at 1.25× input rate and 1-hour cache writes at 2.0× input rate. Most cost trackers lump them together and get the bill wrong by 10–60% on cache-heavy workloads. Pulse reads
cache_creation_5m_tokensandcache_creation_1h_tokensseparately and prices each at its actual rate. - Plan ROI scoring. Anthropic's Console shows API spend. Claude Max users have flat-plan usage that the Console doesn't surface at all. Pulse converts your token volume to an equivalent API bill — so you can answer "is my $200/mo Claude Max plan actually saving me money?" with a real number, not a vibe.
What you get on day one
- Plan ROI — e.g. 10.5× means $2,100 of API value for $200 of subscription
- Per-model breakdown — Opus / Sonnet / Haiku, input / output / cache write 5m / cache write 1h / cache read
- Per-project costs — costs grouped by the project directory Claude Code wrote logs from
- Cost spike alerts — Windows toast when today exceeds 3× your 30-day average
- Multi-currency display — 30+ currencies via daily ECB rates from frankfurter.dev
- CSV export — same format as the interactive demo's sample CSV
How pulse compares to other Claude tools
| Feature | pulse | ClaudeMetrics | Anthropic Console |
|---|---|---|---|
| Reads ~/.claude/projects/*.jsonl locally | ✓ | requires export | — |
| Cache TTL 5m + 1h split pricing | ✓ | approximate | ✓ |
| Plan ROI vs API equivalent | ✓ | — | — |
| Covers Claude Max flat-plan usage | ✓ | ✓ | — (API only) |
| Combines with non-Claude subscriptions | ✓ | — | — |
| Activity tracking (cost per active hour) | ✓ | — | — |
| Runs offline | ✓ | — | — |
| MIT open-source (audit the math) | ✓ | — | — |
| Cost | Free | Free + paid | API price |
The cache-TTL math (why most trackers are wrong)
Anthropic published two cache TTLs (5-minute and 1-hour) at different write prices. Cache writes with the 5-minute TTL cost 1.25× the input price; 1-hour TTL costs 2.0× input. Cache reads are 0.10× input price regardless of TTL.
Older Claude logs only had a single cache_creation_tokens field. Newer logs split it into cache_creation_5m_tokens and cache_creation_1h_tokens. Pulse uses the split fields when present and falls back to the cheaper 5-minute rate when only the bulk field exists — so your ROI is never overstated.
Full pricing table for Opus / Sonnet / Haiku is on /methodology § 3.
Install + first ROI in 5 minutes
- Download pulse for Windows (~80 MB installer) — or run from source on macOS/Linux
- Open the dashboard from the system tray icon
- Complete the 30-second onboarding (currency, optional plan price, activity tracking opt-in)
- Pulse auto-imports your Claude Code logs from
~/.claude/projects/*.jsonl - Plan ROI appears on the Overview tab — refreshed every 6 hours
Try the interactive demo
See the dashboard with fake-but-realistic data before installing — ▶ Open interactive demo. All numbers compute live from the same formulas the real app uses.
⬇ Download pulse · 📐 How ROI is computed · ⚖ Compared to alternatives · ⭐ GitHub