Mission Control Chat
One conversation with your whole AI workforce, how messages route to the right specialist, what tools they use, and how sensitive actions are gated.
What Mission Control Chat is
Chat is the conversation layer over your whole operating system. It's where you talk to the eight specialist agents in plain language: ask a question, hand over a task, or think out loud. The right agent picks it up, works with the same tools and data your team uses, and shows you the result. Markdown renders inline, so answers come back formatted, not as a wall of text.
Chat lives in the Mission Control sidebar. The same agents are reachable on Telegram, WhatsApp, and email, with voice in and voice out for when typing is inconvenient.
How a message is routed
You don't pick an agent from a menu. You write what you need and the message reaches the specialist who owns that area.
| You ask about | Who answers |
|---|---|
| What matters this week, the brief | Atlas (Executive) |
| A stalled deal, a follow-up, lead quality | Forge (Sales) |
| This week's posts, a launch, brand voice | Echo (Marketing) |
| Cash flow, invoices, a late payment | Ledger (Finance) |
| A contract, an NDA, a renewal date | Counsel (Legal) |
| Orders, inventory, a reconciliation | Aria (Operations) |
| Onboarding, reviews, culture | Pulse (People) |
| Access, audit history, anomalies | Cipher (Security) |
When you want a specific voice, call an agent by name ("Forge, which deals are slipping?") and the message goes straight to them.
Tools, not canned replies
An agent in chat is not autocomplete. It reads your live data and acts on it through the same tool surface the rest of the platform uses, 47 tools in total (see The AI workforce). In one exchange an agent might open the actual invoice, check the current pipeline, draft an email against your brand voice, and pull a figure from the ledger. Every tool call is recorded in the trace, so you can see exactly what the agent did.
Approval inside the conversation
When an agent proposes something sensitive, sending mail, moving money, changing a contract, it doesn't just happen. The action surfaces in the thread as a pending card with the draft, the amount, and the recipient, plus an Approve or Review choice. Nothing leaves your workspace until you say so. The gate model is the same one used everywhere else; see Approval gates.
Memory across conversations
Chat is backed by the agents' memory layers, so a conversation isn't amnesiac. Ask a follow-up tomorrow and the agent still knows what you decided today. Memory is per-tenant and never trains a shared model.
Channels
| Channel | Use it for |
|---|---|
| Mission Control | The full chat surface, with traces, approvals, and CRM context beside the thread. |
| Telegram / WhatsApp | The same agents from your phone. Forward a receipt, ask a quick question, approve on the go. |
| Loop an agent into a thread and it replies in context. | |
| Voice | Speak instead of type. The agent transcribes, answers, and can read its reply back. |
Whichever channel you use, the agent, its tools, its memory, and its gates are identical. There's one workforce, not a different bot per surface.
Next
- The AI workforce, the eight agents behind chat and how they run.
- Approval gates, how sensitive actions are held for a human.
- Your first task, hand work to an agent and watch the trace.