Get This Tool
TokenTelemetry
Pricing
- Model
- Free
Summary
Running four AI coding agents locally and having no idea which one burned $200 last Tuesday is the gap TokenTelemetry was built for — a read-only, offline dashboard that reads the logs your agents already write and turns them into cost and trace visibility without touching your codebase.
The tool installs in one curl command, requires no SDK integration, and auto-detects up to 16 agents — Claude Code, Codex, Cursor, Gemini CLI, and a dozen others — by scanning log files on disk. Nothing leaves the machine. The dashboard shows token counts, cost totals, session traces, and reasoning chains per agent. Hermes Agent operators get a dedicated surface covering gateway health, subagent delegation chains, skills, memory state, and cost anomaly detection for silent reasoning-token waste. The ceiling appears fast for teams whose agents do not write logs in a format the parser understands, or who need cross-machine aggregation.
Bottom line: This fits a solo developer or small team running multiple local coding agents who need offline cost accountability — it breaks the moment your agents run on remote infrastructure and you need aggregated telemetry across machines.
Community Performance Report Card
No community ratings yet. Be the first to rate this tool!
Pros
Sign in to edit- Zero-instrumentation setup — reads existing agent log files with no SDK and no code changes, which means you get observability without touching or risking your agent's codebase.
- Fully offline operation with no data upload, so token spend data and agent reasoning traces never leave the machine — critical for teams under data-residency constraints.
- Auto-detects 16 agents from a single install, so you stop maintaining separate ad-hoc scripts per agent just to see which session cost the most.
- Dedicated Hermes Agent dashboard with subagent delegation chains, 38-platform gateway health, and silent reasoning-token waste detection — the only observability surface that treats Hermes as a single agent rather than a collection of disconnected invocations.
- MIT-licensed and self-hosted, which means there is no vendor dependency and no paid-only feature gate blocking core telemetry functionality.
Cons
Sign in to edit- Coverage is bounded by what agents write to local log files — any agent that logs to a remote sink, a cloud service, or a format the parser does not recognize produces zero visibility, and there is no plugin API to extend coverage yourself without forking the repo.
- No cross-machine aggregation exists: if your agents run on a remote server, a VPS, or inside a CI runner, the tool cannot reach those logs, which means teams with remote or distributed agent infrastructure switch to a cloud-native APM tool or OpenTelemetry-based pipeline instead.
- No API surface means you cannot pipe TokenTelemetry data into an existing alerting stack, a cost dashboard, or a downstream data warehouse — teams who need cost anomaly alerts in PagerDuty or Slack have to build that export layer themselves from the raw log files the tool already reads.
About
- Platforms
- macOS, Linux, Windows
- API Available
- No
- Self-Hosted
- Yes
- Last Updated
- 2026-08-14T03:28:22.304Z
Best For
Who it's for
- Developers running multiple local AI coding agents
- Users needing offline cost and behavior analytics
- Hermes Agent operators requiring dedicated observability
What it does well
- Track token usage and costs across multiple coding agents
- Inspect traces and reasoning from autonomous agent sessions
- Monitor Hermes agent health, skills, memory, and source platforms
Integrations
Add notes, reviews, and benchmarks so the next visitor gets a clearer picture.
Compare TokenTelemetry
Spotted incorrect or missing data? Join our community of contributors.
Sign Up to ContributeFrequently Asked Questions
- Is TokenTelemetry free?
- Yes — TokenTelemetry is fully free to use. There is no paid tier.
- Is TokenTelemetry open source?
- Yes. TokenTelemetry is open source.
- Can I self-host TokenTelemetry?
- Yes. TokenTelemetry supports self-hosting on your own infrastructure.
- What platforms does TokenTelemetry support?
- TokenTelemetry is available on: macOS, Linux, Windows.
Best TokenTelemetry alternatives →
Curated lists that include this category
Hidden token costs across local coding agents
Running four AI coding agents locally and having no idea which one burned $200 last Tuesday is the gap TokenTelemetry was built for — a read-only, offline dashboard that reads the logs your agents already write and turns them into cost and trace visibility without touching your codebase.
One-command local install
The tool installs in one curl command, requires no SDK integration, and auto-detects up to 16 agents — Claude Code, Codex, Cursor, Gemini CLI, and a dozen others — by scanning log files on disk. Nothing leaves the machine. The dashboard shows token counts, cost totals, session traces, and reasoning chains per agent. Hermes Agent operators get a dedicated surface covering gateway health, subagent delegation chains, skills, memory state, and cost anomaly detection for silent reasoning-token waste.
Use cases and limits
Use cases include tracking token usage and costs across multiple coding agents, inspecting traces and reasoning from autonomous agent sessions, and monitoring Hermes agent health, skills, memory, and source platforms. Platforms supported are macOS, Linux, and Windows. Integrations cover Claude Code, Codex, Gemini CLI, Cursor, Copilot, Qwen CLI, Vibe, OpenCode, Grok Build, Cline, SmallCode, Pi, Muse Code, Prime Agent, Hermes Agent, and Antigravity. It is free and offers a self-hosted option.
Who it is for / who should skip it
Best for developers running multiple local AI coding agents, users needing offline cost and behavior analytics, and Hermes Agent operators requiring dedicated observability. Teams whose agents do not write logs in a format the parser understands hit the ceiling fast. Skip it if agents log to remote sinks or you need cross-machine aggregation, since no plugin API or API surface exists to extend coverage.