docs/usage/getting-started/resource.mdx
The Resource Library is your personal knowledge hub. Documents uploaded in conversations, AI-generated images, and drafts you create — all collected here. Centralize your knowledge so Agents can answer from your data, not just from general training. Manage and access everything in one place.
The Resource Library provides:
graph LR
A[Upload Documents] --> B[Chunk & Embed]
B --> C[Vector Store]
D[User Query] --> E[Semantic Search]
E --> C
C --> F[Relevant Chunks]
F --> G[Agent Response]
Upload files to your Resource Library through direct upload, conversation uploads, or Notion import.
Documents are:
When queried, the system:
The Agent reads the most relevant chunks and incorporates them into its response — finding content even when the exact words don't match. For example, searching "how to login" can find content about "authentication" or "sign in". </Steps>
Access the Resource Library — Click the Resources icon at the bottom of the left sidebar on the LobeHub main interface.
Organize your resources by theme, project, or type. Separate libraries keep everything structured.
Click New Library in the left panel of the Resource Library page, enter a name, add an optional description, and click Create.
Choose to upload a folder to batch upload all files within it.
Export documents from Notion, then import the Notion ZIP file via the Connect Notion page. Once imported, you can continue editing these documents within LobeHub.
Knowledge Base uses vector-based semantic search. Always resolve pronouns and references to concrete entities.
Why? Vector search works by semantic similarity. Pronouns like "it" or "that" have no semantic meaning on their own and produce poor results.
- "user authentication with OAuth 2.0" ✅
- "login stuff" ❌
- "React component testing with Jest" ✅
- "testing" ❌
- "JSON Web Token authentication" ✅
- "JWT auth" (use after establishing context) ✅
- "How to implement password reset functionality" ✅
- "password reset how" ❌
The recommended approach for getting the best results:
This gives the Agent complete context to answer your question accurately.
For complex queries, search multiple times with different phrasings:
When a topic spans multiple files, read several related files at once to give the Agent comprehensive context.
File chunking — Long documents are split into smaller text segments and vectorized. This helps the Agent understand and retrieve relevant content more accurately.
After chunking:
Select multiple files and open the menu in the top-right corner to perform batch operations: move files to a library, chunk, or delete in bulk.
Supported formats:
File size limit is typically 50 MB per file. For very large documents, consider splitting them before uploading for better search performance.
Search results include relevance scores (0–1). Scores above 0.85 indicate a strong match, 0.7–0.85 indicate moderate relevance worth reviewing, and below 0.7 may not be useful.
Search results include brief excerpts from matched chunks. Review these before reading full files — often the excerpt contains enough information.
When Agents use your knowledge base, best practices include:
Guide knowledge base usage in your Agent's system prompt:
You are a customer support agent with access to our product documentation.
When answering product questions:
1. Always search the knowledge base first
2. Cite specific documents in your responses
3. If unsure, read the full document for context
4. If you can't find info, let the user know
- Verify documents are uploaded and processed (chunked)
- Rephrase query with different terminology
- Use broader, more general search terms
- Check if topic is covered in uploaded docs
- Resolve pronouns to concrete entity names
- Reformulate query with more context
- Try multiple related queries
- Verify document quality and format
- Check if chunking is appropriate
- Use full entity names instead of pronouns
- Break large documents into focused files
- Remove outdated or irrelevant files
- Use specific queries to reduce search scope
- Read only the most relevant files instead of all results
<Card href={'/docs/usage/getting-started/page'} title={'Pages'} />
<Card href={'/docs/usage/getting-started/image-generation'} title={'Image Generation'} /> </Cards>