Captivation Hub's Conversation AI just got a meaningful upgrade. Bots can now learn from a wider range of content — tables, rich text, files — and a new re-ranking layer makes sure the best snippet from your knowledge base ends up powering every reply. Setup is faster, answers are tighter, and you can always see which source the bot pulled from.
What the New Knowledge Sources Support Actually Does
Two big changes are baked into this release. First, your bot can now ingest a much broader set of content types, so spreadsheets, structured documents, and uploaded files are all fair game alongside the URLs and FAQs you've always been able to use. Second, a re-ranking engine sits between the initial search and the AI's reply — it scores every potential match and only sends the highest-quality passages to the LLM. The result is fewer off-the-mark answers and a setup process that takes hours instead of days.
Why It's Worth Paying Attention To
- Tighter, more accurate answers — The re-ranker filters retrieved chunks for semantic relevance before the bot composes its reply, which dramatically reduces hallucinations.
- Roll out faster — Add or swap data sources on the fly. No retraining-from-scratch step.
- Train on whatever you already have — Spreadsheets, rich-text docs, PDFs, slide decks, CSVs — the bot can learn from all of it.
- See where every answer came from — Every bot reply now includes clickable source attribution in the Response Info drawer, so you can trace any answer back to the underlying chunk.
How Re-Ranking Improves Retrieval
Here's what's happening under the hood. After a contact's question hits the system, the initial vector search returns a batch of potentially relevant chunks. Before the LLM ever sees those chunks, the new re-ranker scores each one for how closely it matches the contact's actual question and only forwards the strongest matches. The LLM then writes its answer using just the top passages — leading to replies that are more on-topic and less prone to drift.
Expanded Data Type Support
Your knowledge base is no longer limited to web pages and FAQ pairs. The supported source types now include:
- Tables — Upload structured CSVs to make pricing tables, product specs, schedules, and similar data queryable.
- Rich Text — Paste or write longer-form documentation directly into the knowledge base.
- File Uploads — Drop in PDF, DOC/DOCX, PPT/PPTX, TXT, or CSV files.
- URLs and FAQs — Both still supported and work side-by-side with the new types.
Crystal-Clear Source Attribution
Every bot reply now includes the source it used. Inside the Response Info drawer, you'll find:
- The file or URL name, FAQ label, and timestamp for each source.
- Quick-edit options to correct or replace a source on the spot when you spot something off.
How to Set Up Knowledge Sources
Step 1 — Open or Create a Knowledge Base
Sign in to your sub-account, then click AI Agents in the left-hand menu.
Click into the Knowledge Base tab.
To start fresh, hit + Create Knowledge Base. To update an existing one, click the knowledge base name or the three-dot menu and choose Edit.
Step 2 — Add Knowledge Sources
Click the + Add Source button. Pick from the available types — Tables, Rich Text, File Upload (these are the new options) — and add your content.
Step 3 — Attach the Knowledge Base to a Bot
Inside AI Agents, click Conversation AI, then open the Agent List tab.
To create a new bot, click + Create Bot. To attach the knowledge base to an existing bot, click the bot's name (or use the three-dot menu and choose Edit).
Open the Bot Training tab. Under Knowledge Base, select the one you just created or updated — the new Rich Text, Tables, and Files sources show up here.
Step 4 — Test the Bot Against Your Knowledge Base
Use the Test Your Bot window to throw realistic prompts at the bot. Try edge cases, vague phrasing, and questions that span multiple sources — the answers will tell you what still needs to be tuned.
Step 5 — Inspect Which Knowledge Chunks Powered a Reply
From your account dashboard, head to Conversations and open any thread where the bot has replied.
Next to the AI's message, click AI Response Info. Then expand Knowledge Chunks to see up to three of the chunks that fed into that reply — file name, source, and content all visible.
Frequently Asked Questions
Do I have to retrain the bot after adding a file or table?
Nope. The new content is indexed instantly and the re-ranker pulls from it on the very next query.
Which file types can I upload?
PDF, DOC/DOCX, PPT/PPTX, TXT, and CSV (use CSVs for table-style data).
What's the column limit when I upload a CSV table?
You can import CSVs with up to 500 columns and pick the 20 most relevant ones for the bot to focus on.
Can I mix tables, rich text, URLs, and files in one knowledge base?
Absolutely — and you should. The retrieval pipeline treats every chunk equally up front, then the re-ranker sorts everything together to find the best match across all your sources.
Comments
0 comments
Please sign in to leave a comment.