Token Spend Report
Panopticon tracks every API call, token, and dollar spent across its AI coding pipeline. This interactive report visualizes that data — from model breakdowns to monthly trajectories to the actual codebase evolution video.What’s Inside
This published report is a point-in-time snapshot. For live spend while agents and conversations are running, the dashboard’s Cost Tracking page tracks the same SQLite + JSONL event streams in real time — total cost, event count, budget warnings, a 30-day trend chart, and a per-issue breakdown. The report pulls from two data sources:- SQLite (
~/.panopticon/panopticon.db) — February 2025 through March 2026 - JSONL (
~/.panopticon/costs/events.jsonl) — Canonical from March 31, 2026 onward (includes OpenAI GPT-5.5/5.4)
Key Sections
| Section | What It Shows |
|---|---|
| Hero Stats | Total spend, calls, tokens, and commits |
| Model Breakdown | Top 10 models by cost (Claude, GPT, Kimi, MiniMax) |
| Provider Split | Cost distribution across Anthropic, OpenAI, Kimi, MiniMax |
| Pipeline Stages | Cost by session type: work, plan, review, test, merge |
| Monthly Trajectory | Bar chart of spend over time (with Jan 2026 incomplete-data warning) |
| Top Issues by Spend | Which issues consumed the most tokens |
| Subscription Comparison | What pay-per-use would cost vs. actual subscription spend |
| Cache Savings | Tokens and dollars saved from prompt caching |
| Cost-Reduction Features | RTK, Caveman, TLDR — what’s working and what’s not |
| Codebase Timelapse | Gource visualization of the entire git history |
Beyond Cost: Runtime Metrics
Cost is only half the picture. The dashboard’s Metrics page pairs the spend data above with runtime comparison and performance analytics — useful when deciding which model to route a given role to, since the cheapest model per call isn’t always the cheapest per completed task.Regenerating the Report
The report is generated by the/token-spend-report skill. To update it:
- Run the skill:
/token-spend-report - The skill extracts fresh data from SQLite + JSONL
- Updates the HTML at
docs/token-spend-report/index.html - Optionally regenerates the Gource video
- Commits and pushes to GitHub Pages
~/.claude/skills/token-spend-report/SKILL.md and is synced via pan sync.
GitHub Pages
The report is hosted directly from the repo’sdocs/ folder via GitHub Pages:
main.
Data Freshness
- Last updated: 2026-05-25
- Data through: 2026-05-25
- Git history through: 2026-05-25 (6,463 commits)
- Total tracked calls: 734,117
- Total spend: $54,116