TokenTelemetry is a free, local-first observability dashboard designed to monitor AI coding agents. It provides a read-only interface that tracks token usage, LLM costs, tool calls, session traces, and reasoning steps across a wide variety of coding agents without requiring an SDK, API keys, or cloud connectivity. By reading the log files that these agents already generate, the tool offers a zero-config, privacy-focused solution that ensures your prompts, code, and costs never leave your machine. It was created to help individual developers and teams understand their agent-driven workflows with total transparency.

Functionality is centered on the automated discovery and visualization of agent sessions. Upon installation, the software scans known log locations to index existing agent activity. It then serves a web-based dashboard on localhost, enabling users to monitor live metrics, analyze historical trends, and inspect granular details of individual agent interactions. Because it operates entirely locally, the tool is ideal for environments with strict data privacy requirements.

Some of the key features are:

• Dashboard: A bird's-eye view of your entire agent fleet featuring a live KPI strip, agent distribution charts, and a recent activity feed.
• Agent Compatibility: Built-in support for 13 coding agents, including Claude Code, Codex, Gemini CLI, Cursor, Copilot, Qwen, OpenCode, Vibe, Antigravity, Grok Build, Cline, SmallCode, and Pi.
• Autonomous Support: A dedicated Hermes Agent dashboard that tracks gateway health, scheduled jobs, skills, memory, and 38 source platforms.
• Analytics: Detailed token and cost trends with date-range filters, cache efficiency metrics, and delegation graphs to visualize sub-agent activity.
• Local-First Architecture: Operates without cloud services, signups, or external tracking, keeping sensitive data strictly on your machine.
• Zero-Config Installation: Automatically finds and monitors agent logs without requiring manual instrumentation or changes to your existing codebase.
• Artifact and Trace Replay: Allows users to inspect full session traces, view artifacts, and see reasoning steps with transparent, non-raw summaries.

The tool is operated through a CLI launcher that manages the backend (FastAPI) and frontend (Next.js) services. After a single-line installation, users launch the application from their terminal to start a local web server. The software periodically refreshes to pull new data from agent logs, providing near real-time insights into active sessions. It is also container-friendly, offering support for Docker and Podman to suit different development environments.

Some common use cases include:

• Cost Tracking: An individual developer using multiple coding agents can see exactly how much each AI session costs in real-time.
• Engineering Optimization: A team can compare the efficiency of Claude Code versus Gemini CLI or Codex across their shared projects.
• Agent Behavior Research: Researchers can study LLM agent patterns, tool call execution, and complex reasoning chains for improving prompt engineering.
• Autonomous System Monitoring: A Hermes Agent operator can monitor a Telegram or Discord bot, including jobs and skill-based task delegation across various platforms.