coder-agentmeter

See where your coding-agent tokens go. One local command, no upload.

A unified, local dashboard for multiple coding agents — pick an agent in the header and see its own usage. Auto-detects whatever you have installed.

| Agent | Reads | Tokens | Reasoning | Cache | Est. cost | Rate limit | |---|---|---|---|---|---|---| | Codex | ~/.codex sessions | ✓ | ✓ | ✓ | — | ✓ | | Claude Code | ~/.claude/projects | ✓ | — | ✓ | ✓ | — | | Qwen Code | ~/.qwen/projects/**/chats | ✓ | ✓ | ✓ | — | — | | OpenCode | ~/.local/share/opencode/opencode.db (SQLite) | ✓ | ✓ | ✓ | ✓ | — | | Gemini CLI | ~/.gemini | detected (no per-turn token data) | — | — | — | — |

Run

npx coder-agentmeter

That's it — it opens a dashboard in your browser. Or from a clone:

npm install
npm start

The OpenCode adapter reads a SQLite store via Node's built-in node:sqlite, which needs Node 22.5+. Everything else works on Node 18+. OpenCode is loaded lazily, so older Node still runs fine for the other agents.

Options

coder-agentmeter --port 8080
coder-agentmeter --no-open
coder-agentmeter --codex-home ~/.codex

Per-agent homes can be overridden via env: CODEX_HOME, CLAUDE_HOME, QWEN_HOME, GEMINI_HOME.

Architecture

Each agent is an adapter in src/adapters/ exporting { id, label, mark, accent, capabilities, home, detect, parse }. Every adapter normalizes its raw logs into one shared schema and hands them to aggregate.buildResult(), which produces the daily/model/project/tool/weekday breakdowns and insights. The dashboard reads capabilities to show only the panels an agent supports (rate-limit for Codex, est. cost for Claude, reasoning for Codex/Qwen, …).

To add an agent, drop a new adapter module in src/adapters/, register it in src/adapters/index.js, and the UI picks it up automatically.

• GET /api/sources — all known agents + whether their data is present
• GET /api/data?source=<id> — normalized dashboard data for one agent
• GET /api/refresh?source=<id> — re-parse one agent

Dashboard

• Agent switcher — segmented control of detected agents; per-agent accent/branding.
• KPI strip — total / cached tokens, sessions, and an adaptive third stat (reasoning, est. cost, or output depending on the agent).
• Rate-limit panel (Codex) — live 5-hour and weekly window usage with reset countdowns.
• Overview — daily stacked-token chart, model-share donut, top projects, weekday, tools (all hover-interactive).
• Sessions — sortable, searchable, model-filterable table; click a row for a drilldown drawer.
• Drilldown drawer — every prompt in a session, its turn-by-turn token chart, and tool usage.
• Prompts — most expensive prompts across all sessions.
• Insights — actionable findings (context pressure, reasoning share, tool-heavy sessions, marathon threads, est. spend, rate-limit pressure), each with a concrete "try this".
• Share card — render a 1200×630 PNG of your stats locally (nothing is uploaded).
• Light / dark themes (varna design tokens).

Notes

Token events expose usage (and, for Codex, rate-limit pressure), not a reliable per-token invoice. Costs shown for Claude Code are API-rate estimates, not your subscription bill. Reasoning tokens are reported as a subset of output tokens (total = input + output).