Knowledge sources

The hierarchy

The agent consults sources in this order. Earlier sources outrank later ones.

1. Pinned answers — short, hand-written responses you've written for specific questions.
2. Your docs site — the URLs the crawler indexes.
3. Manually uploaded markdown — internal notes, FAQs, runbooks.
4. Linked GitHub repos — README, docs folder, code comments.
5. Past resolved tickets — successful resolutions get folded back into the corpus.
6. The model's general knowledge — used only when nothing above applies, and the agent flags low confidence.

Adding a docs site

1. From dashboard → Knowledge → Sources, paste your docs URL.
2. Pick crawl depth (default 5). Higher = more pages, longer to index.
3. Optional: add include / exclude path patterns (e.g. exclude /blog/).
4. Click Crawl. Initial crawl typically completes in 5–30 minutes.

Re-indexing

DuggAI re-crawls indexed sites every 24 hours by default. You can trigger an immediate re-crawl from the source row, useful right after a docs update. Clusters attached to old content auto-update when their source page changes.

Pinned answers

When the bot keeps getting one specific question wrong, write a pinned answer. Format:

• Question pattern — natural language phrasing of the question.
• Answer — the markdown reply.

Pinned answers outrank everything else. Use them for:

• Pricing questions (where the truth is in your billing system, not docs).
• Status / availability (e.g. "is feature X live yet").
• Anything sensitive where you want the exact wording every time.

GitHub repos

Connect a GitHub repo and DuggAI indexes README, anything in docs/, and top-level inline comments. Useful for OSS projects where the docs are the repo. Re-syncs on every push.

Manual markdown

Paste markdown directly when you have internal knowledge that isn't (and shouldn't be) public. Common uses: troubleshooting runbooks, escalation playbooks, "known broken" lists.

How retrieval actually works

Each source is chunked, embedded, and stored in a per-project vector index. On every message, the agent retrieves the top chunks across all sources, ranks them by relevance and source priority, and includes the highest-ranked ones in the prompt. The model is explicitly instructed to cite sources and to say "I don't know" when no chunk is sufficient.

What if the agent ignores a source

Usually means the chunk wasn't retrieved — either the source isn't indexed yet, the question doesn't lexically match, or the page is behind auth. Check the conversation detail in the dashboard: it shows which chunks the agent considered.