Name: Harper AI Answer Engine
Brand: Harper

WHY THIS PATTERN

More useful AI, without more AI infrastructure.

Not every question needs a fresh trip through the whole AI pipeline. This build shows how an application on Harper can find the right business context, generate a grounded answer, and reuse that answer when it is safe to do so.

QUALITY

Give users better answers.

Ground responses in the content and operational knowledge your business trusts, instead of leaving the model to answer from general knowledge.

ECONOMICS

Send less work to the model.

Repeated and semantically similar questions reuse validated answers when serving rules allow it, which cuts unnecessary inference.

ARCHITECTURE

Improve what you have.

This Harper application pattern, can sit behind existing search, support, app, and agent experiences.

ECONOMICS AT SCALE

Keep the experience.
Add the answer layer.

Your interface sends Harper a question, and Harper already holds every piece needed to answer it: knowledge and data in its tables, keyword and vector search, an answer cache, routing, grounding, and generation. No separate vector database, cache, or serving layer to wire up. A grounded answer returns to the interface you already own.

Existing experience Site Search

Existing experience Knowledge Base

Existing experience Support Experience

All-in-one Harper

Tables Keyword Search Vector Search Answer Cache Routing Grounding Generation

API Search · Retrieve · Answer

MCP Agents · Tools

Request in Grounded answer or ranked results out

Existing experience Custom Application

Automation Internal Workflows

Agents AI Agent / Model

Existing experience Site Search API

Existing experience Knowledge Base API

Existing experience Support Experience API

All-in-one Harper

Tables Keyword Search Vector Search Answer Cache Routing Grounding Generation

API Search · Retrieve · Answer

MCP Agents · Tools

Request in · Grounded answer or ranked results out

Existing experience Custom Application API

Automation Internal Workflows API / MCP

Agents AI Agent / Model MCP

Built In Capabilities

RETRIEVAL

Hybrid Retrieval

Combine keyword and vector search to find the most relevant evidence for each question, then rank the results into usable context.

LEXICAL + VECTOR + RERANKING

TRUST

Citation Grounding

Generate from retrieved evidence and return supporting sources with the answer, giving users and teams a clear path back to the underlying information.

GROUNDED ANSWERS + SOURCE TRACEABILITY

EFFICIENCY

Safe Semantic Cache

Reuse validated answers when serving rules allow it, avoiding unnecessary model calls while keeping live or entity-specific questions on a fresh path.

FEWER MODEL CALLS + FASTER REPEAT ANSWERS

CONTEXT

Live Business Context

Bring documents, web content, and operational data into the same answer path when the request depends on current business state.

KNOWLEDGE + CURRENT OPERATIONAL DATA

ORCHESTRATION

Model Escalation

Route requests through the lightest path that can answer them, escalating to model inference only when generation or deeper reasoning is needed.

USE INFERENCE WHEN IT ADDS VALUE

PERFORMANCE

On-node Generation

Keep retrieval, application logic, state, and generation close together on Harper, reducing the network hops and coordination required by a stitched stack.

LESS PLUMBING + LOWER OVERHEAD

BUILD IT

Three inputs.
Then run it on Harper.

The retrieval and serving pipeline is not a system you assemble before the first question can be answered.

Connect your knowledge.

Point it at the website, documents, knowledge content, or Harper-hosted data behind the experience.

Choose the model.

Keep model choice separate from the experience. Harper handles the context and answer workflow around it.

Connect your experience.

Send requests through the serving interface that fits your app or agent, and return the result where users already expect it.

THE ARCHITECTURAL DIFFERENCE

The alternative is assembling another AI stack.

A production answer system usually means connecting and operating a collection of specialized components. Harper collapses that answer path into one runtime, and this implementation shows it.

ASSEMBLE IT YOURSELF

One answer layer behind the experience.

Your Knowledge

Your Model

Your Application

Harper

Retrieval, grounding, answer reuse, orchestration, serving, and operational visibility live together in Harper, instead of becoming another integration project.

THE INTERFACES

One answer layer. Multiple ways to ask.

Ask for a complete grounded answer, retrieve ranked context, or expose the same capabilities to an agent. These are the endpoints the implementation exposes on Harper. Clone it and call them yourself.

REQUEST

POST /answer HTTP/1.1
Host: answers.example.com
Content-Type: application/json

{
  "query": "Can I return swimwear without a receipt?"
}

WHAT COMES BACK

{
  "answer": "Yes. Swimwear can be returned within 30 days of purchase as long as it is unworn and the hygiene liner is intact. Without a receipt, store credit can be issued at the current selling price with a valid ID.",

  "sources": [
    {
      "title": "Returns & Exchanges",
      "url": "https://shop.example.com/help/returns"
    },
    {
      "title": "Swimwear Return Policy",
      "url": "https://shop.example.com/help/swimwear"
    }
  ]
}

GROUNDED

2 sources

CACHE

HIT

MODEL CALL

SKIPPED

LATENCY

3 ms

An answer engine, assembled from Harper primitives.

Grounded Answers

Less Inference

Built on Harper

Al answers, inside the search box you already have.

More useful AI, without more AI infrastructure.

Give users better answers.

Send less work to the model.

Improve what you have.

More questions should not mean proportionally more inference.

Keep the experience.
Add the answer layer.

Built In Capabilities

Hybrid Retrieval

Citation Grounding

Safe Semantic Cache

Live Business Context

Model Escalation

On-node Generation

Three inputs.
Then run it on Harper.

Connect your knowledge.

Choose the model.

Connect your experience.

The alternative is assembling another AI stack.

More systems between the question and answer.

One answer layer behind the experience.

One answer layer. Multiple ways to ask.

Deliver grounded answers from Harper.

An answer engine, assembled from Harper primitives.

Grounded Answers

Less Inference

Built on Harper

Al answers, inside the search box you already have.

More useful AI, without more AI infrastructure.

Give users better answers.

Send less work to the model.

Improve what you have.

More questions should not mean proportionally more inference.

Keep the experience. Add the answer layer.

Built In Capabilities

Hybrid Retrieval

Citation Grounding

Safe Semantic Cache

Live Business Context

Model Escalation

On-node Generation

Three inputs.Then run it on Harper.

Connect your knowledge.

Choose the model.

Connect your experience.

The alternative is assembling another AI stack.

More systems between the question and answer.

One answer layer behind the experience.

One answer layer. Multiple ways to ask.

Deliver grounded answers from Harper.

Keep the experience.
Add the answer layer.

Three inputs.
Then run it on Harper.