Skip to main content
You open a new thread and type the thing on your mind — “where did this renewal stall, and who should I nudge?” Your Chief of Staff answers in line: the lead’s last touch, the task nobody picked up, the reply still owed. One question, one place — and the steps it took to get there are right there to expand. That’s the job of this screen: it takes the tab-hunting off your plate, so your attention goes to the call instead of the gathering.

​
What it’s for

Chats is where you talk with your org’s agents. Every conversation is an independent thread — context doesn’t bleed between threads, and you can run dozens in parallel (one per lead, one per project, one for open-ended brainstorming). Unlike messages on a board or comments on a doc, the Chats view is optimized for iterative dialogue: long history, long messages, and actions the agent executed are embedded visually right in the thread.

​
Screen anatomy

Thread list (sidebar)

Threads sorted by last activity. Pin the ones that matter, archive the ones that are done.

Thread header

Thread name, active agent (with model), button to switch persona, and actions (export, archive, delete).

History

Conversation rendered with markdown (code blocks, lists, tables, embeds). Infinite scroll upward.

Composer

Where you type. Supports @mentions (brings in another agent), #leads (links a lead), file drop, and keyboard shortcuts.

​
Multi-persona

Each thread has one active persona — usually your org’s Chief of Staff agent (e.g., “JARVIS”), or a specialist (Scout, Marcus, Athena). You switch persona with:
  • The Persona button in the thread header → pick from the list
  • Mention @scout in your message → Scout joins just for that message (the Chief of Staff remains the thread owner)
  • /persona scout at the start of a message → changes the default persona for the thread from that point forward
Each agent loads its own prompt + memory + tools when it enters. The Chief of Staff reads the full history; specialists read only the excerpt that triggered the call.

​
Anatomy of an agent message

When the agent replies, the message can contain more than plain text:
Rendered markdown — code, lists, tables, links with preview. This is the main content.
When the agent used tools (search, reading docs, looking up a lead), a collapsed Show steps block appears. Expanding it shows each tool called, its parameters, and the result. No mystery about how it reached its answer.
Every response shows its cost in Stars (model + tokens
  • tools used), displayed discreetly in the message footer. See Stars.
When the agent based its response on org documents, a lead, or external research, each claim gets a clickable citation that opens the source. See Brain.
Optional buttons at the end of a response — e.g., “Create task”, “Move lead to next stage”, “Send email”. Clicking executes immediately without you having to repeat yourself.

​
Composer — what you can do

ActionHow
Mention an agent@agent-name — brings in another agent
Link a lead#search → autocomplete; inserts a clickable chip
Link a doc/board&search → autocomplete
Attach a fileDrag-drop, paste, or click the clip icon
Code block``` or three backticks (markdown)
VoiceClick the microphone → speak → automatic transcription
SendEnter sends, Shift+Enter adds a line break

​
Threads — organization

1

New thread

⌘J opens the Chief of Staff agent in a new thread. Or click Chats in the sidebar → + New.Expected result: empty thread with the org’s default persona.
2

Name it (optional)

By default, the thread title comes from the first topic in the conversation (extracted automatically). You can edit it at any time in the header.
3

Pin the important ones

Pin to fix a thread at the top of the list. Useful for the 2–3 active threads in your day.
4

Archive when done

Right-click the thread → Archive. It disappears from the active list but shows up under Archived and stays indexed in search.

​
Slash commands

In the composer, start your message with /:
CommandWhat it does
/persona <name>Changes the thread’s default persona
/clearClears context (does not delete history — just asks the agent to start fresh)
/summarizeRequests a summary of the entire thread
/exportGenerates an exportable markdown version of the thread
/helpLists all available commands

​
Sharing threads

Threads are private by default — only you can see them. You can:
  • Share with a specific member — others see the history and can reply; the agent treats all voices as participants
  • Share via public link (read-only) — generates a URL with an expiration; recipients don’t need to log in
Useful for showing a colleague a great agent exchange, or for archiving a relevant conversation for a retrospective.

​
History — search & analysis

The Chats → Search view searches across all thread history:
  • Substring in message bodies
  • Filter by persona / date / participants
  • Filter by threads that mention lead X / doc Y
Helpful when you remember “someone asked Scout about competitor X last month” and need to find it.

​
Limits & caps

LimitValue
Active threadsNo limit (UI scrolls and indexes)
Messages per threadNo limit — but very long threads (>500 messages) are effectively capped by the model’s context window
Message size~32k characters (if larger, break it into multiple messages)
Attachments25 MB per file, no aggregate cap
Agent spend caps apply — if the org has hit its daily cap, the agent will refuse new requests with “Cap reached. Notify your admin.” See Caps.

​
Audit

Threads are logged:
EventIncludes
chat_thread_createdWho created it + initial persona
chat_message_sentWho sent it (user/agent) + persona + Stars cost
chat_tool_calledWhich tool + parameters (not secret payloads)
chat_sharedWith whom + type (member / public link)
Owners and Admins can inspect these under Settings → Audit.

​
Next steps

Meet the agents

Scout, Marcus, Athena, Twin — who they are and what they do.

Brain

Where the knowledge the agent draws on to respond comes from.

Stars

How much each conversation costs + how to control spending.