How Knowledge Works
When you add knowledge to an assistant, the assistant searches and references it automatically during conversations — you don’t need to prompt it to look things up. Knowledge is layered:Account-Level
Shared account-wide knowledge, turned on or off per assistant with a single switch.
Team
Knowledge available to everyone who uses this assistant.
Personal
Your own private layer, visible only to you.
Adding Knowledge
From an assistant’s Knowledge tab, click Add Knowledge to choose a source:- Choose or Upload Files
- From URL
- Manual Entry
Pick an existing file from your account’s shared store, or upload a new one.Registered file types: PDF, Word (.doc/.docx), Excel (.xlsx), CSV, JSON, HTML, Markdown, plain text, and common image formats (PNG, JPG, GIF, TIFF, BMP, WebP).
The account-level Global Knowledge uploader (see below) only advertises PDF, CSV, TXT and MD, with a 5MB size guideline — a narrower list than what an individual assistant’s Knowledge tab actually accepts. If you’re uploading a large or less common file type, add it through the assistant’s own Knowledge tab rather than the account-level uploader.
Account-Level Global Knowledge
Global knowledge is configured once at the account level (More → Account → Global Assistant Knowledge) as a shared pool of files, URLs, and manual entries. Making it available to a specific assistant is a single toggle — Global Knowledge, on that assistant’s General tab. It’s all-or-nothing: turning it on gives the assistant every account-level source at once. There’s no per-source picker at the assistant level. Global Knowledge is off by default on every assistant.Team vs. Personal Knowledge
Every assistant has two knowledge layers, managed from the assistant’s details view:- Team Knowledge (the Team Settings tab) — shared with everyone who uses the assistant.
- Personal Knowledge (the Personal Settings tab) — private to you, layered on top of team knowledge at runtime.
How Large Files Are Handled
Knowledge is indexed per session, not once up front — each time you start a conversation with the assistant, its knowledge is freshly indexed for that session and discarded afterward. Nothing about the indexing is a one-time setup step. For most files this is invisible. For very large documents, keep things focused: split a large document into smaller, purpose-specific files rather than relying on one huge file for detailed recall, and keep individual files well-structured with clear headings and sections — that consistently improves what the assistant can find. There are working limits worth knowing about: URL fetches are capped in size, and the knowledge available to an assistant in a single session has an overall size cap. If an assistant seems to be missing content from a large source, that’s the most likely reason.Best Practices
Keep knowledge focused and relevant
Keep knowledge focused and relevant
Only add knowledge that’s genuinely useful for the assistant’s role. Irrelevant knowledge dilutes retrieval quality.
Use clear, well-structured documents
Use clear, well-structured documents
Documents with clear headings, sections, and structure are indexed more effectively than dense walls of text.
Use manual entries for short, frequently referenced content
Use manual entries for short, frequently referenced content
For short reference material, manual entries are easy to add — just remember they can’t be edited afterward, only replaced.
Next Steps
General Settings
Configure identity, timeout, and the Global Knowledge switch
Tools & Connectors
Connect to external systems
Capabilities
Image analysis, file generation, and more
Best Practices
Optimise your assistant’s performance