Pulse-Claude-Code-Analytics

See where every Claude Code dollar actually goes.

The open-source Claude Code & OpenAI Codex analytics dashboard + Discord Rich Presence for devs on Claude Pro / Max / Teams and ChatGPT Plus / Pro / Business.Grade your cache A – F, catch runaway sessions before they burn your plan, and copy one-click Fix with Claude Code prompts. Native desktop. 100 % local. Zero telemetry.

Download  ·  What's New  ·  About  ·  Screenshots  ·  Features  ·  Docs  ·  Sponsor

---

• OpenAI Codex support — flip the provider toggle and Pulse reads Codex sessions too. Full GPT-5 family pricing including GPT-5.5 ($5 / $30 per Mtok), 400 K context, and reasoning-effort detection.
• Claude Opus 4.8 — pricing, 1M-context GA, and inflated-tokenizer detection.
• Fast mode, priced right on both sides — Claude Opus 4.8 Fast at 2×; Codex /fast at 2.5× (GPT-5.5) and 2× (GPT-5.4). Each fast turn is flagged with a ⚡ marker in Sessions and Discord presence.
• Per-session Context Window — a session selector, per-session token usage, and tiered compaction hints.
• Reports & Insights no longer hangs — analyzers run off the UI thread; one scan feeds every view.
• Redesigned HTML export — fully offline, on-brand dark theme, smooth charts, fast-vs-standard split.

Download v1.1.0  ·  Full changelog

You pay for Claude Code every month. Maybe $20 on Pro, $100 – $200 on Max, or a seat on Teams. And yet, if someone asked "which of your sessions burned the most context this week?" or "what percentage of your tokens are actually cache hits?" — you probably can't answer.

Pulse answers. It reads Claude Code's own JSONL transcripts (zero-config — just install and launch) and surfaces the things that actually move the needle on your plan:

• A – F cache-health letter grade — trend-weighted, so you see the direction your cache efficiency is heading, not just today's number.
• Opus-4.7-tokenizer aware — Opus 4.7's new tokenizer inflates tokens by up to 35 %; Pulse flags the inflation so you know when you're hitting limits faster than expected.
• 1 M context GA pricing — flat per-token rate across the full 1 M window for Opus 4.6 · Opus 4.7 · Sonnet 4.6 (per Anthropic's official pricing, effective 2026-03-13). Older betas (Sonnet 4 / 4.5, Opus 4 / 4.5) still get 2× input · 1.5× output · 2× cache at > 200 K. Pulse applies the correct math per-model so you compare sessions like-for-like. Plan-level Extra Usage on Pro / Max / Teams is tracked separately.
• Inflection alerts — any session that blows past 2 × your rolling baseline gets flagged with context and a suggested fix.
• Fix-with-Claude prompts — every recommendation has a Copy Fix Prompt button. Paste it into Claude Code. Problem fixed.
• Plan usage limits — live tracking of your current-session, weekly, Sonnet-only, and Extra Usage quotas. Sound alert when Extra Usage spikes.
• Discord Rich Presence — five-tier reasoning effort, live project / model / branch. Your flow state, on your profile.

Works with OpenAI Codex too. Flip the provider toggle and Pulse reads your Codex CLI sessions the same way — per-model pricing for the GPT-5 family including GPT-5.5 ($5 / $30 per Mtok), 400 K context tracking, and reasoning-effort detection. Fast mode (/fast) is priced correctly on both sides: Codex GPT-5.5 bills at 2.5× and GPT-5.4 at 2× the standard rate, and Claude Opus 4.8 Fast bills at 2× — each flagged with a ⚡ marker in Sessions and Discord presence.

Written in Rust + Tauri 2 + Svelte 5. ≈ 12 MB on Windows, ≈ 18 MB on macOS. Cold-starts in under 200 ms. One-click installers for Windows (NSIS + MSI), macOS (DMG — Apple Silicon + Intel), and Linux (deb, rpm, AppImage). Apache-2.0 licensed (attribution required — see NOTICE). The data never leaves your machine.

Dashboard — at-a-glance cost · tokens · cache-hit ratio · plan limits · extra usage · activity heatmap.

Reports & Insights — letter-grade cache health · rule-based recommendations · cost inflection detection · one-click Fix with Claude Code prompts.

Discord Rich Presence — live model · reasoning effort · project · branch · tokens · cost · 5 h / 7 d / Extra Usage, right on your profile.

Windows

irm https://raw.githubusercontent.com/xt0n1-t3ch/Pulse-Claude-Code-Analytics/main/scripts/install.ps1 | iex

Or grab an installer from the latest release:

Asset	Description
Pulse_x.y.z_x64-setup.exe	NSIS installer — recommended
Pulse_x.y.z_x64_en-US.msi	MSI installer

macOS

curl -fsSL https://raw.githubusercontent.com/xt0n1-t3ch/Pulse-Claude-Code-Analytics/main/scripts/install.sh | bash

Asset	Architecture
Pulse_x.y.z_aarch64.dmg	Apple Silicon (M1 / M2 / M3 / M4)
Pulse_x.y.z_x64.dmg	Intel

Linux

curl -fsSL https://raw.githubusercontent.com/xt0n1-t3ch/Pulse-Claude-Code-Analytics/main/scripts/install.sh | bash

Asset	Distro
pulse_x.y.z_amd64.deb	Debian / Ubuntu
pulse-x.y.z-1.x86_64.rpm	Fedora / RHEL
pulse_x.y.z_amd64.AppImage	Any (portable)

From source

git clone https://github.com/xt0n1-t3ch/Pulse-Claude-Code-Analytics.git
cd Pulse-Claude-Code-Analytics
cd frontend && npm install && npm run build && cd ..
cd src-tauri && cargo tauri build

Analytics dashboard

Accurate cost math	Per session · day · model. Published pricing for every Claude — Opus 4.8 (with fast mode), 4.7, 4.6, 4.5, 4.1, 4, 3; Sonnet 4.6, 4.5, 4; Haiku 4.5, 3.5, 3. Speed-aware per-turn, so fast turns price at 2×. Compare like-for-like whatever plan you're on.
A – F cache health grade	Trend-weighted hit ratio — cchubber-style. See your direction, not just your current number.
Model routing insights	Opus / Sonnet / Haiku split + "you could save $X by rerouting N sessions to Sonnet" estimate.
Inflection detection	Any session ≥ 2 × baseline cost-per-session gets flagged with context.
Recommendations engine	Every finding has a Copy Fix Prompt button. Paste it into Claude Code. Problem fixed.
Plan usage limits	Current session · weekly all-models · Sonnet-only · Extra Usage monthly spend. Auto-detects Pro / Max / Teams. Sound alert on Extra Usage spikes.
Heatmap · sparklines · charts	All-local Chart.js. Zero network.
Reports export	Branded HTML + Markdown. One click.
OpenAI Codex support	Full GPT-5 family pricing — GPT-5.5, 5.4, 5.3-Codex, 5.2, 5.1 (+ Codex / max / mini), 5, mini, nano. Fast mode (/fast) surfaced at 2.5× (5.5) / 2× (5.4). One-click provider switch.

Discord Rich Presence

Live fields	Project · git branch · model · reasoning effort · activity status.
Session timer	Elapsed since start. Persists through Discord restarts.
Reasoning tiers	Five tiers (Low / Medium / High / Extra High / Max) — Opus 4.7 native.
Asset resolver	Multi-tier — portal keys → Media Proxy → plain URLs → fallback.
Presets	Minimal · Standard · Full. Customize per-field too.

Privacy & ownership

• 100 % local. Your session data never leaves your machine. No telemetry, no phone-home, no cloud.
• SQLite at ~/.claude/pulse-analytics.db — yours to inspect, export, back up, or delete.
• Apache-2.0 licensed with required attribution. Fork it, audit it, ship your own version — just keep the NOTICE file and credit the original author per the license.

	Pulse	Generic dashboards
Opus 4.7 tokenizer awareness (flags inflated counts)	✓	—
1 M context pricing correctly applied per model (GA flat vs. beta surcharge)	✓	—
A – F cache health grade (trend-weighted)	✓	—
Fix with Claude Code one-click prompts	✓	—
Zero-config — reads JSONL transcripts directly	✓	setup required
Discord Rich Presence	✓	—
Native desktop (Tauri 2 + Rust, no Electron bloat)	✓	—
Open source	Apache-2.0	varies

First launch → install Pulse → launch → it auto-detects your ~/.claude/ folder → keep using Claude Code; sessions stream in live.

Discord Rich Presence → open the Discord tab → pick a preset (Minimal · Standard · Full) or customize fields → toggle on. Custom presence artwork in docs/discord-assets.md.

Reports → open the Reports tab → read your cache grade + recommendations → click Copy Fix Prompt on any item → paste into Claude Code → done. Export HTML / Markdown for your team.

• Linux tray icon — system-tray quick toggles (Discord presence on/off, pause analytics, open dashboard) via Tauri's tray-icon feature. Parity with the Windows tray.
• MCP server inventory — list every MCP server Claude Code has loaded for each session, with its tool count, per-tool invocation frequency, and an estimated token cost per tool call. Helps spot noisy MCPs silently inflating your context.
• Budget-threshold desktop notifications — native OS notification when you cross configurable thresholds (e.g. "80 % of weekly limit", "Extra Usage just passed $150 of your $200 cap"). Replaces eyeballing the dashboard.
• Custom Discord presence templates — save / share named field layouts beyond Minimal · Standard · Full. Export a preset as JSON; import one from a teammate.
• Session replay — step through a past session's prompts and tool-call timeline with the same filters the live dashboard uses. Currently the data is there (JSONL traces) but there's no dedicated UI.
• Smarter cost forecast — weekly-reset-aware projections with a confidence band, instead of today's linear daily-average multiplication.

Track on the project board.

PRs welcome. CONTRIBUTING.md has the dev setup, style guide, and release process. Please read the Code of Conduct first.

If Pulse saves you money (or sanity), sponsor its development:

Every contribution goes toward faster releases, better analyzers, and keeping Pulse free and open-source forever.

Responsible-disclosure policy in SECURITY.md. Report privately via GitHub Security Advisories.

Apache-2.0 © 2026 xt0n1-t3ch. Use Pulse for anything (personal or commercial) — but per the license you must keep the copyright notice and the NOTICE file with original-author attribution in any redistribution or derivative work.

---