Feeding External Information to the AI Agent
The AI agent works not only with the pages and tasks inside a Room, but can also ingest transcripts, notes, emails, chats, PDFs, PPTX files, video files, and more as Smart Intake, so you can use them for search, analysis, and proposals later.
What you can use it for
When sales reps feed in information freely, the AI puts it to use in the following ways.
- Answering and analyzing deal status
- Proposing tasks, Pages, materials, chat messages, and stakeholder updates
- Checking for missing phase-completion criteria
- Extracting customer requests and sales commitments
- Searching and verifying sources later
Adding information from the Room
You can add notes and transcripts from "Intake information" on the Room home.
- Open the Room
- Click "Feed information to the AI" or "Add more information"
- Enter the type of information, the body text, and the processing you need
- After feeding it in, review the AI's organized results and proposals
In a Room with no intake information yet, you'll see "Add meeting notes, memos, or transcripts so the AI can use them for search and analysis."
Attaching from the AI sidebar
You can also attach text and files to the AI agent sidebar in the bottom-right of the Room.
- Open the AI agent
- Attach text or a file from the clip icon
- As needed, drag and drop a file onto the sidebar
- Make a request such as "Summarize this content," "Organize it into a related Page," or "Extract the next actions"
Feeding in a transcript right after a meeting, the body of an email you received from a customer, the original PDF/PPTX files, and Slack/Teams exchanges all together improves the accuracy of later proposals.
The intake information list
"Intake information" on the Room home shows your recent intake.
| Display | Description |
|---|---|
| Title | The subject or file name of the information you fed in |
| Us / Customer | The source of the information |
| Time | The date and time it was fed in |
| Processing | The processing you requested, such as summarization, next-action extraction, or saving to a related Page |
| Status | Analyzing, organized, failed, and so on |
Opening each item lets you review the extracted text, the AI's proposals, and a preview of what will be applied.
Retrying on failure
If the details of an intake item can't be loaded, or if organizing it fails, click "Retry" or "Retry organizing." The original is saved, so you can re-organize it later.
Approving the AI's proposals
From intake information, the AI creates proposals such as the following.
| Proposal | Example |
|---|---|
| Create a task | "Schedule a legal review within 4 days" |
| Update a stakeholder | "Add Director Kobayashi as a candidate decision-maker" |
| Create/append to a Page | "Save the PoC review notes to a related Page" |
| Create material | "Create a one-slide summary explaining ROI" |
| Create a message | "Draft a confirmation request message for the customer" |
When multiple candidates appear, you can check only the items you need and approve them.
- Open the details of the intake information
- Review the preview of what will be applied
- Check only the candidates you need
- Click "Approve selected"
- Read the confirmation text and click once more to confirm
Items that have been applied are shown as "Applied" within the proposal card.
What happens automatically
The following run automatically as safe internal processing.
- Saving the original file
- Extracting text
- Registering it in the search index
- Saving an internal summary
- Drafting AI analysis results
What requires approval
Because the following affect Room content or customer-facing display, they are executed only after approval.
- Creating/appending to a Page
- Creating a Task
- Updating a Stakeholder
- Sending a customer-facing message
- Publishing materials or Pages that customers can see
- Updates that affect the deal phase or CRM
The internal sources the AI referenced are not shown in customer-facing content. The basis is visible only to internal users on the Room-creator side.
Handling meeting transcripts
Because it's difficult to accurately separate speakers from audio or video files alone, transcripts from Meet or Zoom are used by default.
Recommended:
- Enable Google Meet transcription
- Enable Zoom cloud recording and transcription
- Ingest transcripts and metadata via a meeting bot or API
- If you need speaker information, preserve the speaker data from the meeting tool
- Even if you feed in only audio or video files, the original is saved so you can re-organize it later
Upcoming external integrations
As entry points for external integrations, the following ingestion sources are planned.
- Gmail
- Google Meet
- Zoom
- Chrome extension
- Slack / Teams
About saving originals
Originals such as PDFs, PPTX files, recordings, transcripts, and email attachments are saved so you can verify sources later. Deletion and retention periods are managed according to your workspace's retention policy.
Using the AI agent from MCP
When calling the AI agent from external AI tools such as Claude Code or ChatGPT, add the following scopes to your MCP token.
| scope | Use |
|---|---|
agents:read | Answering questions, analysis, and reading |
agents:write | Creating pending actions, approving/rejecting actions, and requesting material creation |
With a read-only token, the AI agent can answer questions and perform analysis, but it does not save pending actions.