Use cases
A private AI chat for your team, with your models, agents and tools
A private AI chat for your team is a chat app like ChatGPT or Claude that runs on your organization’s own model keys and can use your company’s agents and connected tools. In MyDeepAgents it is the chat app at chat.mydeepagents.com: people sign in with their usual account, each person’s chats stay private, and any action that changes something in another app asks first.
Updated · 4 min read
Why give your team a private AI chat?
- One place for every model. Pick from the models of every provider your organization has connected, per message.
- Your keys, your bill. Usage runs on your own provider accounts and shows in your organization’s usage, instead of on personal subscriptions.
- Your work, not just the web. The chat can call the agents you built and the tools you connected: look up a customer, open a ticket, query a database.
- Private by default. A person’s chats are theirs alone; colleagues and admins cannot open them.
What can people do in the chat?
| Feature | What it does |
|---|---|
| Models | Any model from your connected providers, chosen per message |
| Think | Off, on or deep; the model’s reasoning is shown with its answer |
| Web search | Answers with numbered sources, through your search connection |
| Your agents | Chat directly with an agent your organization published |
| Your tools | Use connected tools, MCP servers and outside agents in a conversation |
| Files and images | Up to 5 per message, read by the model |
| Projects | Shared instructions and files for a group of chats |
| Memory | Facts you ask it to remember, which you can review and delete |
| Voice | A hands-free conversation, with spoken approvals |
| Branches | Edit a message or regenerate an answer without losing the first one |
Models
- What it does
- Any model from your connected providers, chosen per message
Think
- What it does
- Off, on or deep; the model’s reasoning is shown with its answer
Web search
- What it does
- Answers with numbered sources, through your search connection
Your agents
- What it does
- Chat directly with an agent your organization published
Your tools
- What it does
- Use connected tools, MCP servers and outside agents in a conversation
Files and images
- What it does
- Up to 5 per message, read by the model
Projects
- What it does
- Shared instructions and files for a group of chats
Memory
- What it does
- Facts you ask it to remember, which you can review and delete
Voice
- What it does
- A hands-free conversation, with spoken approvals
Branches
- What it does
- Edit a message or regenerate an answer without losing the first one
How does the chat use your company’s tools safely?
The chat can use the connections, MCP servers and outside agents of the space a person is working in. Reading runs straight away: look up an order, search tickets, query a database. Anything that writes, such as sending an email or creating a record, stops and shows an approval card with what it is about to do. Nothing happens until the person approves it.
Each message can carry up to 96 tool actions, so pick the tools a conversation needs rather than all of them. The chat runs on the same platform controls as every agent; see AI agent security.
How do you set up the chat for your team?
- In the console, connect at least one model provider with your organization’s key, such as Anthropic, OpenAI or Google.
- For web search, connect a search provider: Tavily, Brave Search, Exa or SerpApi.
- Connect the tools people should be able to use from the chat, and publish any agents you want them to talk to.
- For voice, make sure an OpenAI or Google key is connected: it turns speech into text and back.
- Invite your team. They sign in at chat.mydeepagents.com with the same account they use for the console.
Who can see what people type in the chat?
Only the person who wrote it. A chat, its answers and its approvals are private to their author: to anyone else in the organization, admins included, they do not exist. What the organization does see is usage: how much each model was used, and what it cost.
Memory is visible to its owner too. People can review everything the chat has remembered about them, delete one item, or delete every chat at once.
What does the chat not do yet?
Key points
- One private chat for every connected model, on your organization’s own keys.
- Web search answers with numbered sources through your search connection.
- People can talk to your published agents and use your connected tools.
- Every action that writes asks first, and each person’s chats are private to them.
- Projects, memory, files, voice and branches work as people expect from a modern AI chat.
Questions people also ask
Is the MyDeepAgents chat like ChatGPT?
It works the same way, with models from several providers in one place, and it can use your company’s agents and tools. It runs on your organization’s own model keys.
Can admins read their team’s chats?
No. A person’s chats are private to them. The organization sees usage and cost, not content.
Does the chat send emails or change records on its own?
No. Every action that writes shows an approval card first, and nothing happens until the person approves it.
Which AI models can my team use in the chat?
The models of every provider your organization has connected, chosen per message. See how to choose a model for an AI agent.