claudetokenvampire

Claude TokenVampire

An app that monitors your Claude Code token usage in real time. Anthropic doesn't show you how much of your current 5-hour session quota you've consumed — ClaudeTokenVampire does.

Full documentation lives on the website: gabrielmoraru.com/my-delphi-code/token-vampire

How Anthropic's 5-hour window actually works

Session-based, not sliding. Your 5-hour clock starts on your first message and runs for exactly 5 hours regardless of activity, at which point the counter hard-resets. The next session starts on the first message after that reset. Anthropic uses the word "rolling" in their docs but means "cycles session-to-session", not "continuously sliding".

Source: Anthropic support article 12429409 — "if you hit your limit at 2 PM, your next allocation begins at 7 PM, then 12 AM, and so on."

What it does

It puts you in control of your Claude Code tokens:

• Tracks all billable token types: input, output, cache creation, cache reads
• Shows the current 5-hour session with a per-bucket bar chart from session_start → session_end
• Color-coded bars: green → yellow → red as you approach your limit
• Estimates cost (configurable $/1M token rates)
• Shows cache hit rate and warns when the 5-minute cache gap expires
• Counts down until the session hard-resets (all tokens reset at once, not gradually)
• Tracks the 7-day weekly cap with its own configurable limit and ratio bar
• Top tool calls — a third tab ranks the tools your sessions hit most over the last 7 days, with calls / cost / avg duration
• Runs quietly in the system tray — click the icon to show/hide
• USES 0 TOKENS by default — runs entirely offline, no API calls, no Claude queries (the optional auto-ping feature is opt-in and uses a few tokens per ping)

Features

Data Engine

• Parses all billable token types: input, output, cache creation, cache read
• Sorts entries by timestamp; skips non-assistant entries
• Detects the current 5-hour session: first message where no predecessor exists within 5h; session runs for exactly 5h from there
• Aggregates only entries inside [SessionStart, SessionEnd] — matches what Anthropic counts
• Per-project breakdown, sorted descending by token usage
• Configurable bucket width (2-60 minutes per chart bar)

Computed Stats (per session, global and per-project)

• Total tokens: input + output + cache creation + cache read
• Per-type token breakdown
• Message count (assistant turns) in current session
• Cache hit rate: cache_read / (cache_read + input)
• Cost estimate in USD (four independently configurable $/1M rates)
• Minutes until session hard-reset (SessionEnd - Now)
• Idle minutes since the last message
• Cache gap warning with per-tier detection (5m and 1h shown as two separate gradient bars on the cache-status row)
• Cache tier breakdown: 1h ephemeral vs 5m ephemeral tokens
• Web search and web fetch counts
• 7-day total tokens (true sliding window) with optional weekly cap

All Projects Tab

• Combined stats across all projects
• Five gradient progress bars: token usage, cache hit rate, session-reset countdown, cache 1h warmth, cache 5m warmth (last two share the cache-status row side-by-side)
• Bar chart spanning the current session window (left edge = SessionStart, right edge = SessionEnd)
• Configurable bucket width
• Color-coded bars: green → yellow → orange → red by % of per-slot budget
• Auto-scale blue mode when no limit is configured
• Token value labels above each active bar
• Y-axis with token count labels
• X-axis with hour offsets (start, +1h, +2h, +3h, +4h, end (reset))
• 10% horizontal grid lines; vertical hour-mark grid lines
• Legend (color key or auto-scale note)
• Cache status row: two side-by-side gradient bars (1h tier fills available width, 5m tier fixed-width on the right) plus a short "Xm idle" label — each bar fills as its tier ages toward expiry
• Hot hours warning (13:00-18:59 local time — Anthropic peak-load window, user-reported)
• Detailed tooltips on every stat label

Per Project Tab

• Project list: active projects (with token counts) and inactive known projects (gray, separated)
• Per-project stats: tokens, messages, cache hit rate, cost, expiry, cache status
• Per-project bar chart (same renderer, filtered data)
• Selection preserved across automatic refreshes