Local-first · MIT licensed · live on npm

The Claude Code usage monitor
that respects your machine.

AgentGraphed reads your local Claude Code & Codex CLI logs and turns them into a real dashboard. Timelines, project breakdowns, model usage, cost estimates, one-click resume. No login. No cloud. Nothing leaves your machine.

$npx agentgraphed
Install from npm →Star on GitHub
AgentGraphed dashboard showing 30-day usage chart, KPIs, and today's activity

Built for the way you actually work.

Optimized for the developer who's pair-coding with Claude all day and wants to know where the time and tokens are going.

Every session, every project

AgentGraphed scans your Claude Code and Codex CLI log directories and gives you a real history — searchable, filterable, grouped by day and project.

Cost without surprises

Real cost estimates using LiteLLM's auto-updating pricing data covering 2700+ models. Know exactly what each session ran you.

One-click resume

Open any past session and click resume — AgentGraphed copies `claude --resume <id>` straight to your clipboard.

Optional AI titles & categories

Bring your own Anthropic or OpenAI key (Haiku 4.5 is the default — fractions of a cent per session) for clean past-tense titles like "Fixed Stripe checkout bug" and multi-label classification.

Multi-day session aware

A session that ran across five days appears on all five — tagged Started, Continued, and Closed. No more sessions hiding under one bucket.

Zero configuration

Run `npx agentgraphed` and the dashboard opens. First scan ingests everything in seconds. Re-run to refresh. That's the whole setup.

See it in motion.

Every screen ships in the OSS build. No paywall, no upsell.

Timeline — Every session, grouped by day. Multi-day sessions get STARTED · SPANS Nd / CONTINUED / CLOSED badges so nothing hides on a single bucket.
Timeline
Every session, grouped by day. Multi-day sessions get STARTED · SPANS Nd / CONTINUED / CLOSED badges so nothing hides on a single bucket.
Session detail — Read past conversations in a clean chat-bubble view. Resume in Claude Code or copy a primer for a fresh chat.
Session detail
Read past conversations in a clean chat-bubble view. Resume in Claude Code or copy a primer for a fresh chat.
Projects — Every git repo you've worked in, auto-detected from cwd, ranked by activity, tokens, and cost.
Projects
Every git repo you've worked in, auto-detected from cwd, ranked by activity, tokens, and cost.
Analytics — Sessions per day, provider split, and a model breakdown with cost per model — across 7d / 30d / 90d / all-time.
Analytics
Sessions per day, provider split, and a model breakdown with cost per model — across 7d / 30d / 90d / all-time.
Privacy

Your data never leaves your machine.

AgentGraphed stores everything in a single SQLite file at ~/.agentgraphed/agentgraphed.sqlite. No accounts, no telemetry, no third-party services in the loop.

  • Reads only — never writes to your CLI logs
    Your ~/.claude and ~/.codex directories are read-only inputs.
  • BYO LLM key (optional)
    Titles and classification only run if you add your own Anthropic or OpenAI key, and the traffic goes directly to that provider — never through us.
  • MIT-licensed and open
    Every line is on GitHub. Audit it, fork it, ship it.

Free, forever.

AgentGraphed is MIT-licensed and free for personal use. A hosted multi-developer tier for teams and agencies is coming soon — join the GitHub repo to follow updates.

Solo / OSS
$0
  • ✓ Every feature on this page
  • ✓ Unlimited sessions & projects
  • ✓ Local-only, MIT-licensed
Install with npx
Team (coming soon)
—
  • · Hosted dashboard for your whole team
  • · One source of truth across every dev
  • · Aggregated cost & usage reporting
Watch the repo

Frequently asked.

Does AgentGraphed work without an API key?＋

Yes. Without a key, sessions show the first line of the first prompt as the title. With an Anthropic or OpenAI key, titles get rewritten as clean past-tense summaries and sessions get auto-categorized.

Will it pick up sessions from before today?＋

Yes — the first scan reads every JSONL file in your CLI log directories, all the way back. There's no time limit.

How does it estimate cost?＋

AgentGraphed bundles a snapshot of LiteLLM's pricing data covering 2700+ models. Every session's tokens are multiplied through to a per-session, per-project, and per-model dollar estimate.

Does it work with subscription Claude (Pro/Max)?＋

Yes — the dashboard estimates retail-equivalent cost regardless of how you're billed, so you can see what your usage would cost on the API.

Is there a hosted version?＋

Not yet. A multi-developer hosted tier is on the roadmap for teams and agencies. The solo OSS will always stay free and local-first.