Meeting Context vs. Attachments¶
A meeting can carry two very different kinds of material, exposed through two separate API surfaces. Choosing the right one matters: context is read by the AI agent, attachments are not. This guide explains the distinction so you can pick the correct endpoints.
At a Glance¶
| Meeting Context | Meeting Attachments | |
|---|---|---|
| Purpose | Text-biased material the AI agent should read | Raw files/links kept for reference or download |
| Accepted content | Structured/text formats (JSON, CSV, TSV, XML, HTML, YAML, Markdown, plain text) | Any MIME type (PDFs, images, recordings, arbitrary files) |
| Processing | Converted to clean Markdown, PII-scrubbed, secret-redacted | None — stored byte-for-byte as uploaded |
| Injected into AI prompt? | Yes — visible to the agent | No — never reaches an agent prompt |
| Storage | Sanitized/converted copy | Verbatim original bytes |
| Max upload size | 4.5 MB | 50 MB |
| Typical use | CRM records, spreadsheets, research briefs, text exports | Slide decks, scanned documents, images, call recordings |
| First-party API | /v1/meetings/{id}/context | /v1/meetings/{id}/attachments |
| Partner user API | /v1/partner/user/meetings/{id}/context | /v1/partner/user/meetings/{id}/attachments |
Meeting Context¶
Meeting context is the material you want the AI agent to read and reason over. When you upload a context document it is:
- Converted to canonical Markdown from its source format.
- Sanitized — PII (credit card numbers, SSNs, emails), secrets (API keys, tokens, passwords) and spreadsheet formula-injection patterns are redacted or escaped.
- Injected into the agent's prompt for any chat session scoped to the meeting.
Because context is fed to the agent, it is intentionally text-biased: only structured and text formats are accepted, and binary content is rejected. Use context for anything the agent should actually understand — CRM records, exported spreadsheets, research briefs, prior meeting notes.
Conversion and sanitization run asynchronously after upload. Downloading a context document returns the converted, sanitized copy once preprocessing has produced one; until then it returns the original uploaded bytes. Do not assume a freshly uploaded document has already been scrubbed.
See the AI-Enabled Meeting Prep guide for a full walkthrough of uploading context and conversing with the agent.
Meeting Attachments¶
Meeting attachments are raw artifacts a meeting carries. Any MIME type is accepted and the bytes are stored exactly as uploaded — nothing is normalized, nothing is PII-scrubbed, and the content is never injected into an AI prompt. Attachments are for material humans (or downstream systems) will download and open directly: slide decks, scanned PDFs, images, call recordings, or links to external resources.
Attachments never influence what the AI agent knows about a meeting. If you want the agent to read something, upload it as context instead.
Choosing the Right API¶
- Should the AI agent read it? Upload it as context (
uploadMeetingContext). It must be a text/structured format. - Is it a raw file/link kept for humans to download (or a binary format like a PDF, image, or recording)? Upload it as an attachment (
createMeetingAttachment). - Not sure? Ask whether the agent needs to understand the contents. If yes, it's context; if it's just being stored or shared, it's an attachment.
Attachments are not visible to the AI
Uploading a PDF or image as an attachment will not make its contents available to the agent. Convert it to a supported text format and upload it as context if you need the agent to read it.