The knowledge base is the collection of content that powers a Lyra coach's responses. This page covers what Lyra can connect to, supported file types, and how to curate an effective knowledge base.
On this page: What can Lyra connect to? | How it works | What to include | Supported file types | Fuse learning content
What can Lyra connect to?
Lyra coaches can draw from Fuse communities for knowledge sources.
Best content types for Lyra knowledge base use
|
Format |
Processing quality |
|---|---|
|
PDF (.pdf) |
Richest processing, these are split into chunks using document headers for precise retrieval |
|
Word (.doc, .docx) |
Richest processing, these are split into chunks using document headers for precise retrieval |
Also supported (plain text extraction)
-
Video Transcripts
-
PowerPoint (.ppt, .pptx)
-
Plain text (.txt)
-
HTML
-
Rich Text Format (.rtf)
-
OpenDocument Text (.odt)
-
Word templates (.dotx)
-
PowerPoint slideshows (.ppsx)
-
Macro-enabled formats (.docm, .pptm, .ppsm)
-
Microsoft Publisher (.pub)
Not supported
-
Images (JPG, PNG, GIF, BMP, TIFF)
-
Audio files (MP3, WAV, FLAC)
-
Video files (MP4, AVI, MOV)
-
Spreadsheets (Excel files are not processed for text extraction)
The knowledge base is what makes each coach useful. A coach without relevant knowledge will give generic responses; a coach with well-curated, up-to-date knowledge will give specific, grounded guidance that reflects your organisation's policies, processes, and expertise.
How it works
When you upload a document or connect a content source, the content is processed and indexed so that Lyra can search it semantically — meaning it matches based on meaning, not just keywords. If a user asks "How do I request annual leave?", Lyra can find a document titled "Holiday and absence policy" even though the exact words don't match.
What to include in a knowledge base
The most effective knowledge bases are curated, not comprehensive. Focus on:
-
Content that answers real questions your users are likely to ask
-
Documents that are current and accurate (outdated content produces outdated answers)
-
A mix of conceptual content (policies, frameworks, guides) and practical content (how-to instructions, templates, examples)
Important: Avoid uploading every document you have. A smaller, well-curated knowledge base will outperform a large, unfocused one because the coach is more likely to surface the right content for each question.
Supported file types
|
Category |
Formats |
Processing |
|---|---|---|
|
Best for knowledge base |
PDF (.pdf), Word (.doc, .docx) |
Richest processing — chunked by document headers for precise section-level retrieval |
|
Also supported |
PowerPoint, Plain text, HTML, RTF, ODT, Word templates, PPSX, macro-enabled formats, Publisher |
Plain text extraction |
|
Not supported |
Images, Audio, Video, Spreadsheets (Excel) |
— |
Maximum file size: 500 MB per file.
Tip: If you have important content in a PowerPoint deck, consider also saving it as a PDF or extracting the key points into a Word document. PDFs and Word documents get significantly richer processing, which means better retrieval quality.
Fuse learning content as a knowledge source
In addition to uploaded documents, Lyra can access learning content from your Fuse platform:
-
Articles — Full article body text is indexed for semantic search.
-
Videos — If a video has a transcript, the transcript is indexed as part of Lyra's knowledge base.
-
Questions & Answers — Community-generated Q&A pairs can be indexed, allowing the coach to surface answers from peer knowledge.
-
External links — Links stored in Fuse can be indexed using their title and description metadata.