Open source · Runs on your machine
Your coding agent already logs every turn. This reads those logs and tells you the few changes worth making — each with a number attached.
# 60 days of sample data, then the real dashboard
$ git clone https://github.com/jxxyx-bloop/ai-observatory
$ cd ai-observatory/observatory
$ python3 observe.py demo digest reportPython 3 standard library. No install, no dependencies, no build step.
The output
61% of your spend on time-priced models landed inside a peak window, where the same tokens cost up to twice as much.
Queue the work that doesn't need watching — tests, migrations, doc sweeps — for an off-peak hour.
Cache reuse sits at 38%. Rebuilding context costs roughly 12× what reading it back does.
Keep one session across related tasks instead of restarting. This gap is worth more than any model swap.
Metered, the same work would have cost $412. Nothing here needs fixing.
Stay on the plan. Revisit if your monthly turns drop below 400.
Fifteen checks. Anything worth under $15 a month is demoted, so the top of the list always means something — and healthy usage is reported as healthy.
How it works
Your agent already wrote the logs. We read those files where they are — nothing to install, nothing to switch on.
Tokens, cache, timing and cost — priced at the rate that was actually in force when the work ran.
A ranked list of changes, each with its evidence and what it is worth per month.
No API key, no proxy, no account, no network. Collection costs zero tokens.
Why it's different
DeepSeek and GLM charge by the hour. From UTC+7 to +9, their peak window is your working afternoon.
On an $18 plan, “you spent $412” is fiction. “23× return” isn't.
The 0.1× discount is an Anthropic habit, not a law. Get it wrong and you misprice the biggest number on the page.
IDR, VND, THB, PHP, MYR and more — set against a local day rate, because $412 does not mean one thing everywhere.
Every figure here is public API list price — the ceiling. Enterprise contracts and monthly plans cost less.
Privacy
Nothing leaves your machine unless you edit a file to say so.
Staying current
The dashboard is a file, not a server. The renderer stamps when it ran, and the page does the subtraction itself — offline, with nothing to check in with.
Nothing. A banner that is always on is a banner nobody reads.
Says when it was last built, and that newer sessions are not in these numbers yet.
Says plainly that it is out of date, with the refresh command on a copy button.
Sample data always says so, however freshly it was rendered.
One line. It checks your setup, reads what your coding agents already wrote on this disk, builds your dashboard, puts an icon in your Dock and opens it.
git clone https://github.com/jxxyx-bloop/ai-observatory.git && cd ai-observatory/observatory && python3 observe.py setup
What that command does, step by step →
Still stuck? Run python3 observe.py doctor — it checks every step and prints the exact fix for whichever one failed.