TokenTelemetry is a local observability dashboard for AI coding agents and autonomous agents. It reads the session logs that agents already write on the machine and shows token usage, LLM costs, tool calls, session traces, and reasoning steps for every detected agent in one place. There is no SDK to add to code, no account, and no cloud backend.
Supported coding agents include Claude Code, Gemini CLI, Codex, Cursor, GitHub Copilot, OpenCode, Qwen, Vibe, Antigravity, Grok Build, Cline, SmallCode, Pi Coding Agent, Muse Code, Prime Agent, Qoder, ZCode, and Kimi Code. Agents are auto-detected from their default log locations.
Hermes Agent gets a dedicated dashboard because it runs across CLI sessions, messaging platforms, scheduled jobs, and webhooks rather than in an editor. That view covers 38 source platforms, per-call latency and cache hit rates, subagent cards, skills and memory pages, cron and gateway health, and cost anomaly detection, with pricing that accounts for the provider a model was called through. A Hermes dashboard plugin adds TokenTelemetry as a tab inside Hermes's own web dashboard.
Usage logs, prompts, and costs stay on the machine. The only outbound calls are an optional update check and anonymous product-usage stats, which are on by default and can be switched off. Remote access over a LAN, a tailnet, or an SSH tunnel is opt-in and token-protected.
Features
- Token usage dashboard: tokens in and out per agent, model, and project
- Cost tracking: LLM API cost per session and cumulative over time
- Budgets and alerts: monthly, weekly, or rolling spend or token limits per project and per agent, with alerts at 80% and 100% that never block an agent
- Plan limits: live session, weekly, and monthly quota windows read from the coding-agent logins on the machine, with reset times
- Session traces: a waterfall of system prompt, reasoning, tool calls, responses, and final output
- Tool call analytics: most-called tools with success and failure rates
- Per-project insights: heatmap, activity timeline, and agent leaderboard per codebase, with git worktrees rolled up under the parent repo
- Plan capture: plan-mode outputs from Claude Code and equivalent modes in other agents
- Model analytics: token and cost efficiency compared across models such as GPT, Claude, and Gemini
- Hermes Agent dashboard: autonomous-agent observability across 38 source platforms with gateway and cron health
