VasiHemanth/tokentelemetry

What it solves

TokenTelemetry provides a local, zero-config observability dashboard for AI coding and autonomous agents. It solves the problem of tracking token usage, LLM API costs, and execution traces across multiple different agents (like Claude Code, Cursor, and Gemini CLI) without requiring cloud accounts, SDKs, or manual instrumentation.

How it works

Instead of requiring an API or SDK, TokenTelemetry automatically detects and reads the session log files that AI agents write to the local machine. It then parses these logs to provide real-time analytics on token burn rates, cost tracking, and tool call success rates. It consists of a FastAPI backend that reads the logs and a Next.js frontend for the dashboard.

Who it’s for

Developers and power users who use multiple AI coding agents or the Hermes Agent framework and want to monitor their spend and performance locally and privately.

Highlights

• Multi-Agent Support: Unified view for agents including Claude Code, Gemini CLI, Cursor, GitHub Copilot, and others.
• Local-First Privacy: All data, prompts, and costs stay on the local machine; no cloud signup required.
• Dedicated Hermes Dashboard: Specialized observability for the Hermes Agent, including support for 38 source platforms and subagent tracking.
• Session Traces: Waterfall views of the entire reasoning chain, from system prompts to tool calls and final outputs.
• Observational Budgets: Ability to set spend or token limits per project with alerts at 80% and 100% thresholds.
• Model Analytics: Tools to compare the token efficiency and cost of different LLM models across projects.