Knowledge Bases
Knowledge Bases are the user-facing collections Copilot can use for retrieval-assisted drafting, summarizing, reviewing, and reasoning. A Knowledge Base gives a clear name and purpose to a set of prepared sources, the Copilot profile and governed AI capability that may use them, the retrieval policy that applies, and the latest readiness result.
Use Knowledge Bases when you want members to select approved knowledge by business meaning, such as “HR Policy Knowledge Base”, “Board Pack Drafting Knowledge Base”, or “Vendor Contracts Knowledge Base”, instead of selecting context source IDs or embedding records.
What A Knowledge Base Owns
| Record | Purpose |
|---|---|
| Knowledge Base | The name, purpose, governed AI capability, Copilot profile, retrieval policy, status, readiness summary, source count, and safe evidence refs. |
| Source reference | The prepared AI Context source refs included in the collection, with readiness and evidence summaries. |
| Readiness check | The latest result showing whether sources, profile, retrieval policy, chunks, embeddings, and vector posture are ready for Copilot. |
Knowledge Bases do not own document content, connector payloads, extraction truth, chunks, embeddings, vectors, raw source text, prompts, provider responses, storage keys, or final business records. Those stay with their owner modules.
How To Use It
- Prepare source material through the owning workflow, such as Documents, Enterprise Connectors, and Content Extraction.
- Confirm that the source has an AI Context source and indexing evidence.
- Open AI Copilot > Knowledge > Knowledge Bases.
- Create a Knowledge Base and enter a human-readable name and purpose.
- Choose the governed AI capability and Copilot profile from the configured options.
- Select prepared AI Context sources by label.
- Choose the retrieval policy from the configured options.
- Review the summary and run the readiness check.
- Resolve blockers in the owning area if the Knowledge Base is blocked.
- Use the Knowledge Base in Copilot only when it is ready.
When you choose Use in Copilot on a ready Knowledge Base, Novantra starts an authoring handoff instead of opening a generic AI page. If no editable document target is already known, choose an existing DMS rich-text draft or create a new draft. The DMS editor then opens with the Copilot rail seeded by that Knowledge Base, so the member can draft, rewrite, summarize, improve, or review text from the document editor.
The builder is separate from the details page. Use the builder to create or revise a Knowledge Base. Use the details page to inspect source refs, readiness, evidence, status, last check summary, and the advanced refs that operators may need.
The Preparation step and details view show safe preparation diagnostics such as active chunk count, active embedding count, latest index-run status, indexed/blocked chunk counts, and sanitized blocker reason. A source is not ready for Copilot just because it was extracted earlier: the latest check must still find current source lineage, source-level Content Extraction AI handoff availability, and active chunks that are available for AI handoff. They do not show source text, provider payloads, embeddings, vectors, file paths, or secrets.
For SharePoint-backed knowledge, configure and discover documents in Enterprise Connectors first. The Knowledge Base builder should select prepared AI Context sources after connector discovery, Content Extraction, and AI indexing are ready; it should not ask for SharePoint credentials, connector definition IDs, raw source-ref IDs, or provider JSON.