TokenWatch

Real-time API token usage proxy for Anthropic and OpenAI-compatible endpoints.

TokenWatch sits between your AI tools and upstream APIs, transparently logging token usage, costs, and latency to a local SQLite database. It provides both a CLI and web dashboard for monitoring.

Architecture

Apps (OpenClaw, Claude Code, scripts)
        |
        v
  TokenWatch Proxy (localhost:8877)
        |
   +---------+---------+
   |                   |
   v                   v
Anthropic API     Z.AI / OpenAI API

Installation

pip install -e .

Quick Start

# Start the proxy + dashboard
tokenwatch start

# Configure your apps to point to the proxy:
#   Anthropic: http://localhost:8877/anthropic
#   OpenAI:    http://localhost:8877/openai

Usage

tokenwatch start              # Start proxy (8877) + dashboard (8878)
tokenwatch stats              # Show usage statistics
tokenwatch stats -t 7d        # Stats for last 7 days
tokenwatch tail               # Live tail of requests
tokenwatch status             # Check if proxy is running
tokenwatch reset              # Clear all data

Configuration

Set via environment variables or .env file:

Variable	Default	Description
TOKENWATCH_PROXY_PORT	8877	Proxy listen port
TOKENWATCH_DASHBOARD_PORT	8878	Web dashboard port
TOKENWATCH_DB_PATH	~/.tokenwatch/usage.db	SQLite database path
TOKENWATCH_ANTHROPIC_URL	https://api.anthropic.com	Anthropic upstream
TOKENWATCH_OPENAI_URL	https://api.z.ai	OpenAI-compatible upstream

Web Dashboard

Open http://localhost:8878 for a real-time dashboard with:

• Token usage over time (line chart)
• Tokens by model (doughnut chart)
• Recent requests table with latency and cost
• Timeframe filtering (1h, 24h, 7d, 30d, all)

Cost Tracking

Built-in pricing for:

• Claude Opus 4.6, Sonnet 4.5, Haiku 4.5
• GLM-4.7, GLM-4.7-Flash

---