THE COMPLETE TOOLKIT

Every instrument. One install.

Per-line attribution is the core. The toolkit is what makes it operational — live activity, cost, trends, the menubar, the CLI. All of it ships in the box.

REAL-TIME

Live SSE dashboard

Watch every AI session as it happens. Tokens, cost, current file, current model — streamed live to your browser via Server-Sent Events.

› Per-session cost and token burn
› Current activity table — file by file
› Daily / weekly / monthly trend charts

$ obsly-ai live

localhost:8765 — Claude Code Live Tokens

localhost:8800 — Team Dashboard

USAGE & COST

Team dashboard + LLM proxy

An mitmproxy plugin captures every prompt and token across Claude, Cursor, Codex, and ChatGPT — attributed to the developer, repo, and branch where it happened. The team dashboard rolls it up.

› Cost by repository, developer, model, tool
› Recent activity audit log
› Works with any tool that talks to an LLM API

$ obsly-ai proxy
$ obsly-ai server

TRENDS

Weekly budget burndown

Each week is a spending cycle. The burndown shows where you are vs the linear pace, vs last week, vs the historical baseline. The chart your team checks every Monday.

› Multi-week historical comparison
› Linear pace deviation in real time
› Annotated cycles with context

$ obsly-ai burndown

Weekly Budget Burndown

TERMINAL

CLI: stats and blame, instantly

No browser needed. Two commands cover the day-to-day: see what your AI is producing and who wrote each line.

obsly-ai ai-stats

$ obsly-ai ai-stats

AI Code Statistics · last 30 days
─────────────────────────────────
Total commits     342
AI-attributed     231 (67.5%)
Lines added       12,847
by AI          8,621 (67.1%)
by human       4,226

By agent
  claude-code           5,914
  cursor                2,103
  codex                   604

Durability (30d)  88.4%
Churn (7d)        6.2%

obsly-ai ai-blame src/auth.py

$ obsly-ai ai-blame src/auth.py

  1 human import os
  2 human import hmac
  3 human from datetime import datetime
  4
  5 claude-code class AuthManager:
  6 claude-code def __init__(self, db):
  7 claude-code self.db = db
  8 claude-code def create_token(self, uid):
  9 claude-code return jwt.encode(...)
 10
 11 cursor def validate(self, tok):
 12 cursor return jwt.decode(...)

MACOS

Menubar app

A live AI % indicator in your macOS menubar. Click for today's tokens, cost, durability, and active session — without context-switching to a browser.

$ obsly-ai menubar

Every command in the box

One package. Eleven commands. Zero configuration.

Command	What it does
obsly-ai live	Live SSE dashboard at localhost:8765
obsly-ai server	Team dashboard with cost breakdown at localhost:8800
obsly-ai proxy	Start the LLM traffic proxy (mitmproxy addon)
obsly-ai burndown	Generate the weekly budget burndown chart
obsly-ai report	One-shot HTML usage report
obsly-ai analyze	Text-based token usage summary
obsly-ai menubar	Launch the macOS menubar app
obsly-ai install	Install AI attribution hooks for all detected agents
obsly-ai ai-stats	Show AI authorship statistics for the current repo
obsly-ai ai-blame	Per-line AI attribution for any file in the repo
obsly-ai doctor	Run health checks on the proxy and hook setup