Answers from your docs, with the line they came from.
Developers paste error messages, function names and half-remembered concepts. AACSearch crawls your docs, splits them into sections, and answers with citations — never with an endpoint that doesn't exist.
Queries plain keyword search gets wrong
ECONNRESET socket hang uperror messagehow to paginate list endpointhow-towebhook signature verifyconceptcreateClient optionscode symbolv2 vs v3 authversionsWhere search loses customers today
Whole-page results
Search returns a 5,000-word page; the answer is in one paragraph somewhere in the middle.
Versions mixed up
Results from an old API version rank above the current one.
Hallucinated answers
Generic chatbots invent parameters and endpoints. One wrong answer costs more than ten missing ones.
Tickets for documented things
Support answers questions the docs already cover, because search didn't surface them.
How AACSearch solves it, layer by layer
Each card names the capability behind it. Items marked “in development” are designed and scheduled, not live yet.
Finds the right item from the first letters — typos, synonyms, filters, meaning.
DocSearch crawler
Crawls the docs site, splits pages by headings into sections with anchors and hierarchy.
Hybrid search
Keyword precision for code symbols plus semantic matching for concepts.
Widens the basket: similar, frequently together, personal.
Answers questions a filter can't, citing your own catalogue and docs.
Cited answers
Answers built only from retrieved sections, each claim linked to the source; refuses when the docs don't cover the question.
Safety
Retrieved text is scanned for prompt-injection patterns; answers are checked for grounding in the context.
Answer quality loop
Thumbs up/down on answers and a golden question set scored with nDCG and MRR to track quality over time.
Learns how items, queries and topics connect — and leads the visitor to the goal.
Concept map
Will extract concepts and their relations from the docs and answer broad questions (“how does auth work end to end?”) across many pages.
One visitor, from first query to goal
What the visitor does on the left, what AACSearch does on the right.
- 1
Error
Pastes an error message.
Finds the troubleshooting section that mentions it, not the whole page.
Semantic and hybrid search - 2
Ask
Asks how to verify webhook signatures.
Answers with a code snippet from the docs and links to the section.
AI answers with citations - 3
Go wide
Asks how authentication works end to end.
Summarises across several pages using the concept map.
GraphRAG· in development
Your first two weeks
A typical rollout for this industry. Most teams are live on day one and spend the rest of the time measuring and tuning.
Day 1
Crawl the docs by section.
Day 2
Embed search and the assistant in the docs site.
Week 1
Build a golden set from support questions; review unanswered ones.
Week 2
Compare ticket volume with the month before.
What to measure
The numbers to judge a rollout by. Compare them with an A/B split or a before/after window on the same traffic.
- Ticket deflection
- Support tickets per active user; should fall as answers improve.
- Answer helpfulness
- Share of thumbs-up on assistant answers.
- nDCG on golden questions
- A fixed question set, scored after every change to the docs or ranking.
Connects to
- Sitemap / DocSearch crawl
- Markdown / JSON upload
- REST API
- Embed widget
Capabilities used
Coming next
Questions
Search for Documentation & SaaS, live this week
Connect a feed or a site, embed one script, and measure the difference on your own traffic.