Overview
The knowledge base lets you upload external documents (PDFs, Word files, plain text, Markdown) into ondoki. Uploaded content is extracted, indexed for full-text search, and embedded for semantic search — making it available to the AI chat, MCP server, and search.
Supported File Types
| Format | Extension | Extraction |
|---|
| PDF | .pdf | Text extraction via PyMuPDF |
| Word | .docx | Text extraction via python-docx |
| Plain Text | .txt | Direct text ingestion |
| Markdown | .md | Direct text ingestion |
Uploading Knowledge
Upload files via Knowledge Base in the sidebar or via the API:
Endpoint: POST /api/v1/knowledge/upload
The upload process:
File Upload
File is uploaded and stored on disk.
Text Extraction
Content is extracted from the file format into plain text.
Full-Text Indexing
Extracted text is indexed for keyword search.
Vector Embedding
Content is embedded in the background for semantic search.
Source Types
Knowledge sources are categorized by origin:
| Type | Description |
|---|
UPLOAD | Manually uploaded files |
WEB_CLIP | Content clipped from web pages |
SLACK | Imported from Slack |
MEETING | Meeting notes/transcripts |
GIT_PR | Pull request content from Git |
Currently, UPLOAD is the primary supported source type. Other types are defined in the schema for future integrations.
Knowledge Links
ondoki supports linking resources together with typed relationships:
| Link Type | Description |
|---|
RELATED | General relationship |
DEPENDS_ON | Dependency relationship |
SUPERSEDES | Newer version replaces older |
PART_OF | Component of a larger whole |
Knowledge Graph
The Knowledge Graph page provides a visual representation of relationships between documents, workflows, and knowledge sources. Navigate to it from the sidebar.
Managing Knowledge Sources
List sources: GET /api/v1/knowledge/sources
Delete a source: DELETE /api/v1/knowledge/sources/{source_id}
Deleting a knowledge source removes the file, extracted content, and associated embeddings.
How Knowledge Integrates with AI
Knowledge base content is available to:
- Search — appears in unified search results alongside documents and workflows
- AI Chat — the
rag_search tool queries knowledge embeddings to provide context
- MCP — external AI agents can search and access knowledge content
