Every active community quietly produces a library of answers — buried in threads, replies, docs, and shared files. MateFlow's AI Copilot turns that into something members can actually find, answering questions from your own source material with inline citations instead of generic guesses.
The knowledge base is managed under Admin → AI, where you'll also find quotas, usage stats, and question logs. AI Ask answers by retrieving from two things you control: knowledge sources you add deliberately, and your community content.
Add knowledge sources
Open Admin → AI → Knowledge Base and choose Add Source. There are three source types:
- File Upload — PDF, Word, Excel, PowerPoint, Markdown, Text, CSV, HTML, RTF, EPUB, or JSON, plus images (PNG/JPG/WebP/GIF/TIFF) that are auto-OCR'd. Files are used only for parsing and aren't stored in your media library.
- URL Import — crawl a website via Cloudflare Browser Rendering. Set entry URLs, allowed domains, whether to include subdomains, crawl depth, and optional include/exclude path patterns. Sitemap scanning is supported and respect robots.txt is always on.
- Q&A — hand-write question/answer pairs. These index instantly on save, making them the fastest way to fill a content gap when the AI keeps missing something.
Every source has a priority (higher values win when sources overlap) and an enabled toggle. Prioritize the sources you trust most and keep them current.
Parsing, chunking, and indexing
Under Advanced settings, each non-Q&A source exposes a parser mode (Auto, Basic, or Advanced high-fidelity Markdown conversion) and a chunking mode that controls how documents are split before embedding: Auto, Fixed window, Parent/child (retrieve small chunks, expand to a larger parent), or Q&A pairs. Defaults work for most sources.
File and URL sources index as background jobs you can watch from the source list. You can run a full index, an incremental crawl, or a reindex (clear and rebuild from scratch), batch these across many sources, view job history, and cancel a running job. Status badges show Indexed, Indexing, Queued, Failed, or Cancelled, along with pages fetched, docs, and chunks.
Bring in community discussions
Your best answers already live in posts. Under Admin → AI → Community Content for AI, enable retrieval so the copilot searches community discussions alongside your knowledge sources. You control:
- Scope — all spaces, a whitelist, or a blacklist. Only content spaces (Post, Blog, Q&A) are eligible; chat, media, podcast, and course spaces are always excluded.
- Priority spaces — give official documentation or key spaces a relevance boost (1.2x to 3.0x; 1.5x recommended).
- Advanced retrieval knobs like Top K (how many posts to retrieve) and a score threshold.
The more your community talks, the more the copilot has to draw on.
Grounded answers with citations
Answers are grounded rather than generic: each AI response cites its sources, rendered as numbered inline markers and source cards that show the type (Post, Course, Event, Blog, Space, or External), title, a short summary, and a relevance score. Members can click through to the original, so they can verify every claim.
Shape the assistant and watch the gaps
Under Admin → AI → Settings you can give the assistant a display name, avatar, greeting, and up to six suggested starter prompts shown on the AI chat page. AI runs either platform-managed or with your own provider key (BYOK), and you can toggle individual capabilities (completion, actions, and ask) and per-member daily limits.
Finally, use the Hit Logs to review real member questions with confidence scores. Low-confidence or unanswered questions reveal content gaps — the perfect prompt to add a Q&A entry or a new source, letting the copilot deflect repetitive questions so your team focuses on the ones that need a human.
Next steps
- See the full AI Copilot feature overview.
- Organize the spaces the AI draws from with Spaces.
- Point URL and file sources at your existing Courses and docs.
- Check which AI capabilities and quotas each tier includes on the pricing page.