Train With Your Knowledge
Training lets you feed your company's existing knowledge into your employees so they can answer questions and make decisions based on your real data.
TL;DR
Train from four kinds of source: a connected App (Notion, Slack, HubSpot, and 60+ more), a Website (crawl any URL and its sub-pages), a File (upload PDFs, Office docs, spreadsheets), or plain Text (paste notes, a policy, an FAQ). Pick what to include and the system builds a knowledge base your employees search during every conversation. Training is company-wide, so all employees share the same knowledge. Train once, everyone benefits.
Four Ways to Train
You do not need a connected app to train. Pick whichever source fits the knowledge you have:
| Source | What it is | Best for |
|---|---|---|
| Apps | Connect a tool and pull its content (Notion, Slack, HubSpot, and 60+ more) | Knowledge that already lives in your stack |
| Websites | Paste a URL and crawl the page and its sub-pages | Public docs, help centers, your own marketing site |
| Files | Upload PDFs, Office docs, spreadsheets, or text files | One-off documents not stored in a connected app |
| Text | Paste raw text: notes, a policy, an FAQ, anything | Quick facts you want your team to know right now |
All four feed the same shared knowledge base and behave identically once trained: a session with live progress, deduplication, and automatic retrieval during every conversation.
How It Works
Training pulls content from your chosen source and builds a searchable knowledge base that employees automatically use during every conversation.
The pipeline:
| Step | What happens |
|---|---|
| 1. Connect a source | Connect an app from the Tools tab |
| 2. Click Train | Hit the Train button on the connected app's card |
| 3. Pick sources | A source picker opens showing available items (pages, channels, folders, projects). Select what to include |
| 4. Fetch content | The system pulls documents via API. Each document becomes searchable text |
| 5. Build knowledge | The system extracts key information and builds a searchable knowledge base |
| 6. Ready to use | Employees automatically search this knowledge during every conversation |
Session tracking. Every training run creates a session with real-time progress. You can see how many documents were found, how many were processed, and whether any failed. Pipeline steps (searching, building graph, complete) appear in a timeline.
Queuing. You can start a new training run while one is already running. The new run is queued and starts automatically when the current one finishes.
Supported Apps
Beyond Websites, Files, and Text, training works with 65+ connected apps across 10 categories. We add more regularly, so the list inside the product is always the source of truth: open Training to see everything currently available and connect what you use.
Source picker. Most apps let you choose exactly what to include. Notion shows pages, Slack shows channels, Jira shows projects, HubSpot shows object types (contacts, deals, companies). The picker adapts to whatever content structure the app uses. If no picker appears, the system trains on all accessible content.
Do not see your app? Open Training in the product to check the live list, which is always ahead of this page. If it is still missing, it may not expose its content for training yet.
Where to Find It
Open the Training tab on your Company Dashboard. Four source tiles sit at the top:
- Apps opens the app browser to connect or retrain a tool
- Websites asks for a URL to crawl, including its sub-pages
- Files lets you upload PDFs, Office docs, spreadsheets, or text
- Text gives you a box to paste notes, a policy, or an FAQ
You can also train a connected app straight from its card in an employee's Tools tab. The Train button appears when the app is connected and trainable.
What You Can Do
| Action | How |
|---|---|
| Train from a connected app | Click the Train button on the app's card in the Tools tab |
| Select specific sources | Use the source picker to choose pages, channels, folders, projects, or object types |
| Train everything | If no source picker appears, the system trains on all accessible content |
| Queue multiple runs | Click Train again while one is running. It queues automatically |
| View training progress | Watch the session timeline: documents found, processed, skipped, errors |
| Retrain when content changes | Click Retrain to pull fresh content. Unchanged documents are skipped automatically |
How to Set It Up
- Open Training on the Company Dashboard and pick a source tile: Apps, Websites, Files, or Text
- Give it the source. Connect and pick an app, paste a URL, upload files, or paste text
- Pick what to include. For apps, select pages, channels, projects, or objects in the picker. Or train on everything
- Wait for completion. Watch the session timeline for progress. Training runs in the background and does not block chat
- Done. All employees now have access to that knowledge automatically
Tips and Tricks
- Start with your most important docs. Train on your core documentation first, then add more sources over time
- Use the source picker. Selecting specific sources gives you cleaner, more relevant knowledge than training on everything
- Retrain when content changes. If your source documents are updated, run training again. The system skips unchanged content and only processes new material
- Training does not block work. Employees can chat and work normally while training runs in the background
- Combine multiple sources. Train on Notion for your docs, Slack for team context, HubSpot for client data, and Jira for project context. Each run adds to the same shared knowledge base
- CRM training is powerful. Training on HubSpot, Salesforce, or Pipedrive gives every employee instant knowledge about your clients, deals, and pipeline without needing to ask
Good to Know
- Company-wide, not per-employee. Training applies to the entire organization. All employees share the same knowledge. You do not need to train each employee separately
- Knowledge appears automatically. After training, employees search the knowledge base on every message. You do not need to tell them to use it
- Sessions are tracked. Every training run creates a session with full document-level detail: what was ingested, what was skipped (duplicates), and what failed
- Queuing is automatic. If you start a new training run while one is active, it queues and starts when the current one finishes. No runs are ever rejected
- Source picker varies by app. Notion shows pages, Slack shows channels, Gmail shows labels, Jira shows projects, HubSpot shows object types. The picker adapts to whatever content structure the app uses
- Duplicate detection. Documents are deduplicated automatically. If the content has not changed since the last training run, it is skipped
- Not every app supports training. Only apps with text-rich content (documents, messages, issues, CRM records) support training. Analytics tools, payment processors, and utility services are for real-time use, not training
Frequently Asked Questions
Q: Do I need to connect an app to train? A: No. Apps are just one of four sources. You can crawl a Website, upload a File, or paste Text directly, with nothing to connect.
Q: Where do I see the full list of supported apps? A: Open the Training tab in the product. The in-app list is always the source of truth and ahead of this page. We add new apps regularly.
Q: Does training work per employee or per company? A: Per company. All employees share the same knowledge. Train once, everyone benefits.
Q: How long does training take? A: It depends on how many documents you are training on. A handful of Notion pages takes under a minute. A large Google Drive with hundreds of files takes longer. Training runs in the background and does not block anything.
Q: Can I train on multiple sources at once? A: You train one connected app at a time, but you can queue multiple runs. Start training on Notion, then immediately start training on Google Docs. The second run queues and starts when the first finishes.
Q: What happens if I train on the same content twice? A: Documents are deduplicated automatically. If the content has not changed, it is skipped. Only new or modified content is processed.
Q: Do I need to tell employees to use the trained knowledge? A: No. Employees automatically search the knowledge base during every conversation. The results are injected into their context silently.
Q: Does training cost credits? A: The fetching and ingestion phase is very low cost. The knowledge-building step uses minimal processing. The cost is negligible compared to regular chat.
Q: Can I untrain or remove specific content? A: Not yet from the UI. If you trained on wrong data, retrain with the correct sources selected. A selective "forget" feature is planned.