Skip to content

Document Hub

The Document Hub tool group gives the agent the ability to search for documents stored in the OMD Document Hub using semantic (vector) search.


How it works

Documents are pre-processed and stored as embeddings in ChromaDB (the vector store). When the agent performs a document search:

  1. The user's query is embedded into the same vector space.
  2. The most semantically similar document passages are retrieved.
  3. The agent presents the relevant excerpts with source references.
  4. The conversation continues for follow-up questions.

Available tools

Tool Description
search_documents Search for relevant passages in the tenant's document store using a natural-language query
search_documentation Search docs.optimizemyday.com for relevant OMD product documentation

Document ingestion

Before documents can be searched, they must be ingested into the vector store. This is done via the /documents API endpoint (admin operation):

POST /documents/ingest

The agent fetches documents from the OMD Document Hub, splits them into chunks, generates embeddings, and stores them in the customer_documents ChromaDB collection.


Example interactions

User: "Find the instructions for handling hazardous material on-site."

The agent: 1. Calls search_documents with the query. 2. Returns the most relevant passages from the document store. 3. Cites the source document and page.


User: "What does the OMD documentation say about configuring time windows?"

The agent: 1. Calls search_documentation with the query. 2. Returns relevant excerpts from docs.optimizemyday.com.


Collections

ChromaDB Collection Content
customer_documents Tenant-specific documents from the OMD Document Hub
documentation_documents OMD product documentation pages
conversation_summaries Embeddings of past conversation summaries (used for memory retrieval)
memories User and organisation memory facts