Chat with your agents, not a generic assistant.
It works like the AI chat everyone already knows. The difference is the other side: not a general-purpose assistant, but your agents — connected to your systems, each with its own computer to work on, under rules you set.
spend against budget by month:- • Northwind alone is 38% of the overspend, all in expedited freight.
- • Contoso's
Q3renewal repriced mid-quarter and nobody caught it.
Create an agent. It's already in the chat.
Rolling a bot out through a chat tool means registering an app, requesting scopes, and managing its tokens. Here, chat is built into the platform — the moment an agent exists, everyone it is shared with can message it.
Created at 14:02, first conversation at 14:03. Nothing was deployed in between.
Every agent has its own computer
An agent here is not a text box over a model. Each one works on its own machine — with a filesystem for your files, tools it can install, and internet access that follows rules you set.
So the reply to “compare these contracts” can be the comparison — unpacked, computed, and attached — not a suggestion for how to do it yourself.
You can see it working
Agents show their status in the thread list — running, pending, finished — the way colleagues show presence. Running means running on the platform: close the laptop and the answer is waiting when you are back.
And the record keeps: behind every message is the full run that produced it, one click away.
One chat for your team and your agents
There is no switching between the chat for people and the tool for AI. A thread holds colleagues, an agent, or both — ask together, watch the same answer land, hand a conversation over.
The conversation belongs to the thread, not to somebody's account.
Which suppliers drove the Q3 overspend?
Include the freight surcharges this time.
Four suppliers — freight included, chart attached.
See it on your own cluster.
Thirty minutes, your infrastructure, your stack. Or skip the call — it is one Helm release onto a cluster you already run.