Package Exports

This package does not declare an exports field, so the exports above have been automatically detected and optimized by JSPM instead. If any package subpath is missing, it is recommended to post an issue to the original package (@agentdx/agentdx) to support the "exports" field. If that is not possible, create a JSPM override to customize the exports field for this package.

Readme

AgentDX

See what your AI coding agents are actually doing.

A small, local-first tool built by a developer who wanted an honest picture of how the different agentic coding tools — Claude Code, Codex, Cursor, Aider, Cline — really behave: where the tokens go, which one ships, where the errors pile up. It reads logs your agents already write and shows you the reality. Runs entirely on your machine; nothing ever leaves.

Install

npx @agentdx/agentdx

Or install once and call it by name:

npm install -g @agentdx/agentdx
agentdx

Homebrew: coming soon.

Screenshots

One npx @agentdx/agentdx and the dashboard opens on your own machine — no account, no upload.

Overview — the flight recorder at a glance

AgentDX overview dashboard

Insights — judgements, not just numbers ("which agent actually ships?")

AgentDX insights page

Models & Tokens — spend, cache hit rate, input vs output

AgentDX models and tokens page

Sessions — every coding-agent session, searchable and drillable

AgentDX sessions page

What It Does

AgentDX is a free tool that reads the local logs your AI coding agents already write, stores everything in a single SQLite database on your machine, and serves a dashboard so you can see tokens burned, tools called, models used, and errors hit across every agent — and judge which one actually ships. One command, no accounts, no uploads, no telemetry. Nothing ever leaves your computer.

Supported Agents

Agent	Source	Status
Claude Code	`~/.claude/projects/`	Stable
Cursor	`~/.cursor/projects/`	Stable
Codex	`~/.codex/`	Stable
Aider	Local chat history files	Stable
Cline	VS Code extension storage	Stable

AgentDX reads local logs and artifacts. It never intercepts live traffic or modifies agent behavior.

Quick Start

# 1. Collect session data from all installed agents
npx @agentdx/agentdx collect

# 2. Start the dashboard
npx @agentdx/agentdx serve

# Or do both in one command
npx @agentdx/agentdx

The dashboard opens at http://localhost:3002. (After npm install -g @agentdx/agentdx you can drop the npx @agentdx/ prefix and just run agentdx.)

CLI Commands

Shown with npx @agentdx/agentdx; once installed globally, just use agentdx.

Command	Description
`agentdx`	Collect data from all agents, then start the dashboard
`agentdx collect`	Collect session data only (no dashboard)
`agentdx serve`	Start the dashboard only (no collection)
`agentdx watch`	Collect, serve, and re-collect on an interval (default: 5 min)
`agentdx status`	Show database stats: sessions, messages, tool calls, tokens, models
`agentdx import <file>`	Import sessions from a portable JSONL export

Privacy

AgentDX is 100% local. It makes no network requests of any kind — no accounts, no uploads, no telemetry, no analytics. Your sessions, prompts, code, tokens, and repo names never leave your machine. The only thing it writes is a single SQLite file at ~/.agentdx/agentdx.db, which you fully control and can delete anytime.

Full details in PRIVACY.md.

License

MIT