Every AI session.
Every agent.
One bill.
Cross-agent AI cost visibility for every developer.
Claude Code · OpenAI Codex · OpenCode · GitHub Copilot. All four captured automatically. Real tokens, real cost, real model attribution — per session, per project. No agent cooperation needed.
Free · Open Source · MIT License · Node.js 18+
Without Worktale
- ✗ You see the Anthropic/OpenAI invoice — no idea which project burned it
- ✗ No idea how many tokens your longest session ate
- ✗ Can’t tell whether Opus is actually worth it vs Sonnet on routine work
- ✗ AI session context vanishes when the chat closes
- ✗ No record of the decisions behind AI-generated code
With Worktale
- ✓ Cost per session, per model, per project — automatic on all four major agents
- ✓ Input, output, and cache token counts captured automatically
- ✓ Agent narrates intent, decisions, and trade-offs after every commit
- ✓ Claude Code, Codex, OpenCode, Copilot — plus rules for Cursor, Cline, Windsurf
- ✓ Local-first — your cost data stays on your machine
What a tracked session looks like
Narrative per commit, session metrics at the end — automatic on Claude Code.
commit a3f8c21
fix: resolve webhook race condition with idempotency keys
Tracked down a race condition where two webhook deliveries for the same payment event were both being processed. Added idempotency key check before processing. Chose database-level unique constraint over application-level locking because the webhook handler runs across multiple workers.
commit b7e2d45
feat: add retry backoff for failed payment webhooks
Implemented exponential backoff with jitter for failed webhook retries. Max 5 attempts over ~30 minutes. Considered a dead-letter queue but decided it was overkill for current volume. Will revisit if we exceed 10k events/day.
Session end — captured automatically:
Three commands. That’s it.
npm install -g worktale
One command to install globally. Works on macOS, Linux, and Windows. No dependencies.
worktale init
Run once per repo. Scans your entire git history and installs a silent post-commit hook. Done in seconds.
worktale digest
End of day, get a clean summary of everything you shipped. Add notes. Build your record.
Then connect your AI agent. Claude Code:
/plugin marketplace add worktale/worktale-plugin
—
all platforms →
CLI dashboard · Desktop app.
Same local database. Browse in the terminal TUI, or open the desktop app for charts. Pick your tool.
Commits
7
Lines +
+1,204
Lines -
-389
Files
23
Time
4h 12m
🔥 STREAK ━━━━━━━━━━━━━━━━━━━━━ 14 days
📈 BEST ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 31 days
• 2m ago fix: resolve auth token expiry edge case
• 47m ago feat: add retry logic to payment processor
DAILY LOG · ◀ Monday, March 3, 2025 ▶
8 commits · +892 / -201 lines · 14 files
Your Notes
Spent the morning tracking down a race condition in the payment webhook handler. Classic.
09:14 fix: webhook race condition (idempotency key)
10:02 fix: add mutex to payment handler
14:17 feat: implement retry backoff for failed webhooks
HISTORY · 2024 – 2025
Mon ░▒▒█████░░████████░███████░░█████
Wed ░████░░█████████░████████░░██████
Fri ███░░░███████████░█████████░░░███████
Total commits 4,847
Longest streak 31 days
Lines written 892,044
Days active 287
AI SESSIONS · last 30 days
Sessions
142
Cost
$47.83
Tokens
8.4M
Active
38h
Cost/day
▁▃▂▅▄▇█▆▄▂▃▅▇▆▄▃▁▂▄▆█▇▅▃▂▁▃▅▄▂ peak $4.12
AGENTS
claude-code 84
codex 32
opencode 17
copilot 9
TOP MODELS
claude-opus-4-7 71x
gpt-5 28x
sonnet-4-6 23x
Desktop app — same data, interactive charts. Windows, macOS, Linux.
0
network requests
0
accounts required
0
telemetry / tracking
100%
local SQLite storage
Works with every major AI coding agent
Claude Code
Native plugin
Copilot CLI
Plugin + hooks
Codex CLI
Skill
Cursor
Rules
Cline
Rules
Windsurf
Rules