How Azaris works
Connect your tools, tell it how you work once, then ask for things in the app already open on your phone, while an isolated machine, a credential vault, and a written trail run underneath.
Step 1: Connect your tools
Each connector is a click, and you choose what it can see. Start with the pair that carries most of your day, email and calendar, then add channels (Telegram, Slack, Discord) and the rest of your stack as the work calls for it. Disconnect any of them at any time; access is revoked, not hidden.
AI is included: your plan comes with AI credits every month, with top-ups if you need more. Prefer your own? Connect your ChatGPT plan or an OpenAI or Claude key, and your provider bills that usage instead.
Step 2: It learns how you work
There is no training run and your data is never used to train a model. The mechanism is simpler and more inspectable: instructions you write, files you hand it, and a memory it keeps between conversations. When it learns something about you, the change shows in its memory history, and you can undo anything that isn't right.
Step 3: Ask in your channel
Message it in Telegram, Slack, or Discord, talk to it by voice, or use the web app. It's the same agent with the same memory everywhere. It shows the steps it takes, cites what it read, and checks with you before big decisions.
What runs underneath
- Your own machine. Every customer's agent runs on a dedicated machine with its own persistent storage, created for you, destroyed with you.
- Keys in a vault. Credentials live encrypted in a vault; the plaintext is injected only into your machine, at start.
- A sandboxed browser. Web browsing happens on a separate isolated machine, away from your data, and you can watch it live.
- A written trail. Every step is logged in plain language, so you can audit an afternoon or a month.
The longer version of this list, with the reasoning behind it, is on the security page, and the comparison with running an agent runtime yourself is in OpenClaw vs Hermes.