Everything in MeghaOS starts with a request you type into the chat. There’s no command syntax to memorize; you describe what you want and the agent figures out whether to render a widget, run a tool, kick off a workflow, or just answer.

​
The four things a request can become

​
Fast-path commands

A handful of inputs skip the LLM entirely and return instantly:

​
Writing good requests

Be concrete

“a habit tracker for water, sleep, and reading” beats “a tracker”. The agent seeds the widget with your specifics.

Name the data

“stock chart for NVDA” or “weather in Kyoto”: concrete entities route straight to the right live widget.

Chain freely

“search for the top AI papers this week and email me the list” becomes a two-step workflow automatically.

Ask to explain

“show me how a 4-stroke engine works” yields an animated SVG diagram, not a wall of text.

​
Memory makes it personal

The agent remembers facts across sessions. Once you’ve said “I’m based in Tokyo” or “my work email is sam@acme.com”, later requests like “what’s the weather?” or “email the team” use that context automatically. See Memory.

​
Approvals for risky actions

Tools that could be destructive, or any plugin the agent wrote for itself, may ask for confirmation before running. You’ll see a confirmation card; approve or decline it. Trusted built-in plugins (is_safe = True) run without prompting. See the Plugin System.

​
Conversations & history

Chats are grouped into sessions you can revisit or delete:

​
Other ways to talk to it

MeghaOS also listens outside the desktop. With the right tokens configured, you can message it from Telegram, Slack, or Discord, and it can receive Google Chat / Teams webhooks. Replies come back through the same channel.