Data sources

A source is anything your agent should be able to quote when answering a question. At training time we read every active source, split it into searchable chunks, and store embeddings for retrieval at chat time. That means:

• Adding a source by itself does nothing — you have to retrain for the agent to start using it
• You can toggle sources on/off without deleting them, useful for A/B testing
• Storage is shared across all sources for an agent, capped by your tier

Drag a PDF onto the upload card, or click to open the file picker. Each file caps at 10 MB; multi-file uploads work in one drop.

When to use it

• Product manuals, datasheets, white papers
• Internal docs you'd otherwise paste into a Notion page
• Anything you've already curated for human staff

Two modes: paste a URL and Chatmount crawls it, or paste a list of specific URLs to ingest only those pages. The crawler respects robots.txt and a configurable depth.

Crawler tips

• Start with your /help or /docs subpath rather than the root domain — much higher signal-to-noise
• After a crawl, deselect marketing pages individually if the agent ends up sounding salesy
• Use the per-page Retrain button on the link detail view to refresh a single page

Pages are grouped by domain in the link list — handy when you’ve fed in URLs from multiple sources.

A textarea. Paste content directly. Best for short, structured knowledge that doesn’t live anywhere else yet — refund policies, escalation rules, brand voice notes, glossary terms.

Question + answer pairs you author manually. The agent treats these as authoritative — when a user’s question matches one of your questions, the answer comes back nearly verbatim.

Reach for Q&A when

• An answer needs to be exact (legal, pricing, eligibility)
• The agent gets the same question wrong repeatedly — easier to overrule with a Q&A than to rewrite the source
• You're plugging holes flagged in Suggestions

Click any row in the list to open the detail modal — same flow as adding a new pair, just edits the existing one.

OAuth into your Notion workspace, pick which pages and databases the agent should have access to. We re-sync on every retrain so updates in Notion flow through.

Pro tier. Ingest historical support tickets from Zendesk or Salesforce so the agent learns from past resolutions. The connector flow lives in the Actions tab — you authenticate once, then pick which queues / cases get pulled in.

Pro tier. Surfaces gaps in your knowledge base — questions users asked that the agent answered with low confidence, or sources that contradict each other. Each suggestion comes with a one-click draft of a Q&A snippet you can accept to fix the gap immediately.

Total storage across all sources is capped by your tier. The bar in the right rail tints amber at 70% and red at 90%. Two ways to free up space: delete unused sources, or upgrade.

• Free — small cap, suitable for a couple of PDFs or a tight Q&A set
• Plus — comfortable for a small product's full doc set
• Pro — supports large knowledge bases, ticket history, and Suggestions