01
Perceive
The incoming WhatsApp or web message arrives with conversation state — what was already asked, captured, or retrieved.
AI engine
AI that routes, retrieves, and stays grounded.
The agentic core behind every AskHandle agent. It classifies each message, chooses the right tool — document search, data search, live browsing, or your briefing — then answers, captures, or hands off. Customers get a reply in seconds, on WhatsApp and the web.
Agentic routingDocument searchGrounded answers
AI engineAgentic core
PerceiveRouteRetrieveActRemember
AI Answer
Long briefing in context. Replies in seconds.
Live web browsingOn
Document Search
KeywordMeaningCross-language
Retrieved from Guest handbook.pdf, page 4. Answer grounded in that passage.
Data Search
SKUPriceURL
LF-100016$249.00/pro
Values taken from the matching row. Not estimated, not rewritten.
Agentic core
How the agent thinks on every message.
Search is one step. Agentic AI is the loop around it: the agent decides what to do, uses the right tool, keeps state, and can act — not only chat.
02
Route
The Router classifies intent on every turn, not only the first. When the goal changes, the path changes with it.
03
Retrieve
Document Search, Data Search, or live browsing run only if needed. Evidence is fetched before a reply is written.
04
Act
The agent answers, captures structured fields, books a time, or hands off to a person. Chat is one outcome, not the only one.
05
Remember
Verified rows and lead fields stay available for follow-ups, so “which of those is cheapest?” does not start from scratch.
Route
The Router sends each message to the right intelligence.
You do not pick one model of answering and hope. The agent classifies the message, then only the matching nodes run.
Policy or handbook question
Document Search
Price, SKU, or catalog lookup
Data Search
Conversation, live facts, or your briefing
AI Answer
Capabilities
Four ways the AI engine produces a better answer.
A generic chatbot treats every question the same. AskHandle gives the agent the right kind of intelligence for the job.
AI Answer
Long context, held in one briefing.
Teach the agent who it is, how it should speak, and what it knows. FAQs, policies, product notes, and house rules stay in context for every reply. The briefing can hold a large body of stable knowledge, up to 200,000 words, without splitting it across prompts.
See the skills you can turn onAI Answer briefingReplies in seconds
You are the Northshore assistant. Help with plans, booking, and house rules. Never invent prices. If you are unsure, offer to connect them with a person.
Live web browsingOn
Live browsing
Answers that can look up what is true right now.
Turn on live web browsing when the question depends on the current world: weather, hours, news, rates, local places. The agent searches while it answers, instead of relying on a frozen snapshot of the internet.
Looking up live results
hours for Northshore Thursday
northshore.co/hours
Open Thursday 9:00 AM – 6:00 PM
Thursday 2:00 PM is open. I can hold it and send a confirmation.
Document Search
Find the right passage, then answer from it.
Policies, manuals, contracts, and training files are retrieved before a reply is written. Hybrid search combines exact keywords for codes and IDs with semantic matching for meaning. Ask in one language and retrieve from documents written in another.
Retrieved passageGuest handbook.pdf · p. 4
Check-in is after 3:00 PM. Early check-in depends on housekeeping. Quiet hours begin at 10:00 PM.
KeywordMeaningCross-language
Data Search
Prices, SKUs, and URLs from the row, not a guess.
Spreadsheets and catalogs are searched as structured records. The agent identifies the matching rows and fields, then answers with those values verbatim. Identifiers, prices, and links are not paraphrased or invented from general knowledge.
Matching row
SKUPriceURL
LF-100016$249.00/pro
Taken from the sheet. Not estimated, not rewritten.
FAQ
Common questions
What is agentic AI in AskHandle?+
The agent does not treat every message the same. It classifies intent, chooses a tool — Document Search, Data Search, live browsing, a briefing, lead capture, or human handoff — keeps state for follow-ups, and only then replies. That loop is the AI engine.
Is Document Search the same as RAG?+
Yes in the everyday sense: retrieve the relevant passage, then generate under evidence constraints. Data Search does the same job for spreadsheets — rows and fields, not passages. Responses are anchored to what was retrieved, which reduces hallucination risk compared to pure generation.
What does long context mean on AI Answer?+
The AI Answer briefing can hold up to 200,000 words. That is where you put role, tone, rules, and stable company knowledge so the agent does not have to rediscover it every turn.
When should I turn on live browsing?+
When the answer depends on something that changes outside your files: weather, opening hours, news, or local recommendations. Leave it off when the agent should stay inside your briefing and connected search nodes.
Why are Document Search and Data Search separate?+
Documents need passage retrieval. Spreadsheets need row and field anchors. One generic file search mixes those jobs and is how prices and SKUs get invented. Split retrieval keeps each answer under the right contract.
Will the agent invent a price or SKU?+
Data Search is built so it does not. Matching records are retrieved first, and those field values are used as written. If nothing in the sheet answers the question, the agent says it does not have that information.
Does this AI run on WhatsApp and the website?+
Yes. You assemble the agent once. The same AI Answer, browsing, document search, and data search run on WhatsApp, the web chat widget, and a shareable chat page.
Put this agentic core in your agent.
Assemble the loop in the builder. Go live on WhatsApp or the web chat widget.