A living changelog of the platform behind Agents.Deployed. It's been in active, near-daily development for months — here's the journey, from a blank server to what you have today. Read top-to-bottom for the whole story, or jump to the latest release at the end.
2026.06.14 — your team grew (a researcher and a
markets analyst), and the assistant now reacts to your messages. Full details at the bottom.A bare Linux server and a raw, unconfigured AI engine. The idea: a private assistant that's genuinely yours — living 24/7 on your own machine, not a rented seat in someone else's cloud.
It builds up memory from your conversations over time — your preferences, facts you mention, recurring details — instead of starting fresh each time.
Fixed a bug that could silently drop tasks and reminders it was tracking for you.
A live "working on it…" indicator while it searches or thinks, so you always know it's busy — and every reply now comes through, even when it's using a tool behind the scenes.
When you ask it to buy or book something, it always shows the final total and asks you to confirm first — including which payment method to use (a saved card, Google Pay, or a card you give it with spending limits). No surprise charges.
Reliably uses your Calendar, Gmail, and Contacts when asked — no more wrongly claiming it can't.
Ask how something works or what a setting does, and it answers accurately about your setup instead of guessing.
Keyboard (incl. non-Latin layouts) and copy/paste now work properly in the browser desktop, plus a cleaner login screen and tidied workspace.
The system backs itself up several times a day and automatically repairs its database if it ever hiccups — so a rare technical glitch fixes itself on a restart, with no manual help.
The multilingual cleanup now also covers replies that come from the behind-the-scenes specialist helpers — so stray garbled characters never slip through, no matter which part of the system answered.
In the rare event the AI gets stuck repeating itself, it now catches that instantly, hands you the useful part of the answer with a short note, and stops — instead of flooding you with a wall of repeated text. A built-in length cap keeps any single reply (and your usage) sensibly bounded.
Your assistant now delegates to a researcher and a markets analyst. For a deep, multi-source question the researcher digs through the sources and hands back a short, cited summary instead of a wall of text. The markets analyst tracks your portfolio and watchlist, remembers why you hold each position, and answers "how's BTC / my portfolio?" with real numbers — always with a clear not financial advice note. You still talk to one bot; they work behind the scenes.
The assistant now adds emoji reactions to your messages when it fits — a quick, natural acknowledgement, instead of only ever replying in text.
Still building. New improvements ship continuously — this page grows with them.