Knowledge Base
Manage indexed workspace sources, ask questions across all knowledge or one source, and keep ingest and index health visible while you curate reusable material.
Usage
Use Knowledge Base to keep useful material available across your workspace instead of searching through old sessions one by one.
This is useful for recurring reference docs, research packs, previous meetings, and other material you want to reuse.
The Knowledge Base page is now chat-first: the default right panel is an embedded chat that can answer across all indexed sources or stay scoped to one selected source.
Knowledge Base is also the source layer behind chatbot answers that cite reusable workspace material instead of only the current session.
If you want chatbot conversations to answer workspace knowledge questions consistently, Knowledge Base is the system you build first.
Ask across all knowledge
Use this when you want chatbot to search the whole indexed workspace instead of one source.
At least one source has finished ingest and indexing.
You are on the Knowledge Base page with no individual source selected.
Open Knowledge Base from the sidebar, keep All knowledge selected at the top of the left list, and ask your question in the embedded chat on the right.
Ask one source only
Use this when you want answers limited to a meeting, upload, or session bundle instead of the full workspace.
The source exists in the left Knowledge list.
The source has reached a searchable ready state or you accept that pending ingest may limit results.
Click a source in the left list. The right panel stays on Chat and scopes retrieval to that source. Use Ask this source from Source details if you opened the details view first.
Review ingest and index health
Use this when a source exists but search or chat still misses content, or when you need to confirm indexing finished.
You can open the source in the left Knowledge list.
Open Source details with the panel icon on a source row. Check Ingest stage, Index consistency, Pipeline, and Embedding. Re-ingest from the source row if the source changed or indexing looks incomplete.
Before you start
Knowledge Base availability depends on the current account plan and workspace configuration.
Non-session knowledge uploads can require a paid plan depending on the backend setting.
Uploaded knowledge sources still go through validation and ingest before they become searchable.
Session sources, uploaded resources, and edited knowledge documents can appear separately while the index catches up.
The left panel lists sources under Knowledge. Each row shows status, document count, ingest stage while processing, and freshness when the index may be out of date.
The right panel has two views: Chat and Source details. Chat is the default landing view.
Chatbot answers about reusable workspace knowledge depend on the indexed Knowledge Base content, not only on raw files existing somewhere in the product.
Knowledge settings control which completed session types are added to Knowledge Base automatically, so they directly affect what future chatbot retrieval can see without manual curation.
Facts, action items, and evidence spans are extracted by AI models, so they can contain mistakes, omissions, or incorrect wording.
Users should review and edit extracted facts, action items, and evidence spans before treating them as reliable sources for search or chatbot answers.
Step-by-step instructions
Open the chat-first Knowledge Base workspace
The page is organized as a left source list and a right working panel. Chat is the default experience.
Open Knowledge Base from the sidebar
Use the Knowledge Base entry in the workbench sidebar.
Start from All knowledge
The top row in the left list is All knowledge. Keep it selected to ask chatbot across every indexed source in the workspace.
Filter the source list when needed
Use Search knowledge in the left panel to filter sources by title or type without leaving the page.
Resize panels on desktop
Drag the divider between the left source list and right panel to give more room to chat or source details.
Add sources and auto-ingest rules
Add sessions or uploads
Use Add sessions to turn completed sessions into knowledge sources, or Add resource to upload a supported file bundle.
Review Knowledge settings
Open Knowledge settings to choose which completed session types are auto-added to Knowledge Base, such as meeting recordings, browser recordings, uploads, net videos, or AI voiceover outputs.
Save the auto-ingest rules
After changing the toggles, save the settings so future completed sessions of those types are added to Knowledge Base automatically.
Wait for ingest and indexing
VoxStudio creates the source first, then extracts, chunks, and indexes it. The source row shows pending, ingesting, ready, or failed status until indexing completes.
Use scoped chat and source details
Select a source for scoped chat
Click a source in the left list to scope chat to that source. Click All knowledge again to return to workspace-wide answers.
Open Source details for curation
Use the panel icon on a source row, or Ask this source from the details header, to switch between chat and the extracted content view.
Review ingest metadata
In Source details, inspect Ingest stage, Index consistency, Pipeline, and Embedding so you can tell whether chunk indexing finished and which extractor or embedding versions were used.
Edit extracted knowledge
Use the Documents, Facts, Action items, and Evidence tabs to review and correct AI-extracted content before relying on it in chat or search.
Re-ingest when content changes
Use the re-ingest action on a source row after major edits, failed ingest, or stale freshness so search and chat can catch up to the latest content.
Tips and examples
Think of Knowledge Base as the retrieval layer for chatbot, not only as a storage folder.
A good mental model is: source enters KB, KB is ingested and indexed, chatbot retrieves from that index, then the answer shows citations.
Use All knowledge for broad workspace questions and scoped source chat when you already know which meeting, upload, or bundle should answer the question.
Index consistency is the fastest health check: if indexed chunks are lower than total chunks, wait for ingest to finish or run re-ingest.
Use Knowledge settings to keep the index aligned with your real workflow instead of relying only on manual add actions.
Example: if your team wants chatbot to answer questions about internal meetings by default, keep Meeting and Record Transcribe enabled in Knowledge settings so those completed sessions flow into KB automatically.
Treat AI-extracted knowledge as a draft that needs review, not as a final source of truth.
Use clear titles so you can find sources quickly later, especially when filtering the left list.
Prefer one focused source per topic or project when possible. Search quality is usually better than with large mixed bundles.
If a source changes materially, re-ingest it instead of assuming search results have updated already.
Refresh source detail when you are watching ingest status, because indexing can finish after the source record already exists.
Example: if you upload a customer onboarding SOP into KB, chatbot can answer questions like "What is the escalation path for onboarding delays?" only after that SOP has finished ingesting.
Example: if AI extracts the wrong owner for an action item from a meeting, that wrong owner can show up later in chatbot answers unless a user corrects it in Knowledge Base.
Common problems and fixes
Why can’t I add a non-session source?
Your current plan may not allow free-form knowledge uploads, or the feature may be disabled for your environment.
Why is the source visible but not searchable?
The ingest job may still be pending, failed, or waiting for a re-ingest after an edit. Check the source status row and Index consistency in Source details.
Why does Index consistency show fewer indexed chunks than total chunks?
Indexing may still be running, a re-ingest may be in progress, or the last ingest run did not finish cleanly. Refresh the source and re-ingest if the counts stay mismatched.
Why did a file upload fail?
The source may be too large, unsupported, or missing a valid uploaded object when the ingest completion step runs.
Why does search or chat miss expected content?
The source may not have ingested correctly, the wording may differ from your query, or the document needs to be refreshed and re-ingested. Check ingest status and Index consistency in Source details, then re-ingest if needed.
Why can chatbot not answer a KB question even though I uploaded the file?
Uploading alone is not enough. The source must finish Knowledge Base ingest and indexing before chatbot retrieval can use it.
Why did a finished session not show up in Knowledge Base automatically?
Its session type may be disabled in Knowledge settings, or the auto-ingest flow may not have completed yet.
Why is a fact, action item, or evidence span wrong?
Knowledge extraction is performed by AI models. The model can misread speakers, timing, ownership, or meaning, so users should review and edit extracted facts, action items, and evidence spans.
FAQ
What is the difference between All knowledge and selecting one source?
All knowledge lets chatbot retrieve across every indexed source in the workspace. Selecting one source limits chat to that source only.
Can I use a session as a knowledge source?
Yes. Session-derived knowledge is one of the main ways to build reusable internal context.
Can I edit a knowledge document later?
Yes. Edited knowledge documents are re-ingested so search can catch up to the latest content.
Does deleting a source remove it from search?
Yes, once the delete and index update flow completes.
What does Knowledge settings actually change?
It decides which completed session types are pushed into Knowledge Base automatically. That changes the future retrieval coverage chatbot can rely on without manual source creation.
Should I enable every source type in Knowledge settings?
Not always. Enable the types that produce reusable knowledge for your team. Leaving noisy or low-value source types off can keep the index cleaner and improve chatbot relevance.
How exactly does Knowledge Base affect chatbot answers?
Knowledge Base provides the indexed sources that chatbot retrieves from when you ask workspace knowledge questions. Better KB coverage usually leads to more grounded chatbot answers.
Can chatbot answer KB questions without Knowledge Base?
Not reliably for reusable workspace knowledge. Without KB, chatbot may only have the current session context or other directly available evidence.
Should I trust AI-extracted facts, action items, or evidence without checking them?
No. AI extraction can be useful, but it can also be wrong. Review and edit important facts, action items, and evidence spans before relying on them.
Is Knowledge Base the same as Shared Sessions?
No. Knowledge Base is for reusable internal source material and chatbot retrieval. Shared Sessions are for distributing a public session view.