Coding-agent telemetry
Telemetry, usage, and cost analysis for CLI coding agents such as Claude Code, Codex CLI, Gemini CLI, and OpenCode, whether you use them day to day or are evaluating them.
- Claude Code monitoring - Claude Code’s built-in OpenTelemetry export, turned on with
CLAUDE_CODE_ENABLE_TELEMETRY=1and an OTLP exporter setting, which sends session, token, and estimated-cost metrics and prompt, API-request, and tool-result events to any OTLP endpoint, plus per-prompt traces when the betaCLAUDE_CODE_ENHANCED_TELEMETRY_BETA=1is also set, with prompt and tool content redacted unless you opt in; Claude Code itself is proprietary. - AI Observer - Single-binary OTLP receiver with a DuckDB store and web dashboard for Claude Code, Codex CLI, Gemini CLI, Copilot, and OpenCode telemetry, which can also import past session files and price their token use; MIT.
- AgentsView - Local-first session search, analytics, and token-use statistics for the Claude Code, Codex, and other coding-agent session archives on your machine; it analyses the agents you use, not apps you build.
- ccusage - CLI that reads the local session logs of Claude Code, Codex, OpenCode, and other coding-agent CLIs and reports token use and estimated cost by day, week, month, or session, and for Claude Code by 5-hour billing block; MIT.
- Verovo - Local CLI, formerly AgentMeasure, whose
checkcommand reads Codex and Claude Code session logs and reports duplicate records, retry chains, and repeated tool failures, marking what the logs cannot decide as UNPROVABLE; most of its other commands audit AI vendor bills against usage records; MIT. - ranwhat - CLI that reads the session history that Claude Code and 29 other coding agents keep on your machine and reports, after the fact, the risky actions they ran (credential file reads, destructive Git commands and deletes, package publishing) and the secrets left in the transcripts; MIT, with an optional paid hosted plan for syncing findings across machines.