tokenforge-cli
v1.0.0
Published
Predictive AI spend control — see where your tokens go, forecast what's coming, and stop budget overruns
Maintainers
Readme
TokenForge
Predictive AI spend control. See where your tokens go, forecast what's coming, and stop budget overruns before they happen.
Fork of codeburn by AgentSeal, extended with predictive spend intelligence and a premium visual dashboard.
What it does
- Anomaly detection — flags spending spikes with natural language summaries
- Burn rate forecasting — 7d/30d averages, trend direction, budget ETA
- Pre-flight estimates — predicts next session cost range from historical data
- Budget tracking — real-time gauge with overage alerts
- Cost heatmap — GitHub-style calendar view of daily spend
- Sankey flow — visual breakdown of spend by provider, model, project
- Model efficiency — cost-per-edit and one-shot rate per model
- Savings tracking — shows money saved by routing to free/cheaper models
- Data confidence — trust layer that flags stale or unpriced data
Install
git clone https://github.com/fortuenti/tokenforge.git
cd tokenforge
npm install
npm run build
npm linkUsage
tokenforge web # open the dashboard
tokenforge status # quick CLI summary
tokenforge budget 500 # set monthly budgetDashboard
The dashboard runs locally at http://localhost:53975 with live SSE updates. All data stays on your machine.
Supports Claude Code, Cursor, Codex, Windsurf, Kimi, and other AI coding tools.
Requirements
- Node.js >= 22.13.0
License
MIT — see LICENSE
Original work copyright (c) 2026 AgentSeal. Fork modifications by Fortuenti.
