Home/AI capabilities/Knowledge and memory
Knowledge and memory

Let AI answer based on your information instead of guessing based on common sense

Product information, quotations, and FAQs are put into a dedicated knowledge base, and AI will give priority to these contents when replying. Where did the customer chat last time and what requirements did they raise? Write it down across sessions so you can pick it up next time.

What it provides

  • Knowledge basesYour information, the first reference source when AI replies
  • Long-term memoryCustomers’ historical needs continue across days and sessions
  • Confirmation before warehousingThe content that affects the external caliber will be reviewed by you before taking effect.
  • Exportable recordsHistorical conversations and customer files can be exported and retained.
Why it exists

You've probably run into these

AI answers product details entirely by making up information, and the customer reveals the details as soon as he asks about the parameters.

Put product information, parameter tables, and quotations into the exclusive knowledge base. When AI replies, it will take priority from here. It should not fill in content that is not in the knowledge base.

The customer's needs mentioned last week will be asked again this week in another session.

Cross-session long-term memory retains the customer's stated needs, preferences, and chat progress. The next conversation will be connected directly without the customer having to repeat it.

The data is scattered in group chats, network disks, and several people's computers, and no one synchronizes it after changes.

The knowledge base and material library are stored centrally, and pictures, videos, and files can be uploaded. Change one place and all the accounts and conversations that reference it will change.

The customer chatted for three months and wanted to review but could not find the complete record.

Historical sessions and conversations are saved by customer, and customer files can be exported. In the data warehouse, you can review by time and dimension without scrolling through the chat window.

Capabilities

What it can do

Divided into three layers: editable knowledge, referenceable materials, and automatically accumulated memories. The first two layers are maintained by you, and the third layer is generated during the conversation.

Knowledge bases

Create a new knowledge base and maintain it separately by product line or market. Entries can be added, deleted or modified, and you can return to the knowledge base list to switch at any time.

Resource Base·Knowledge Base

The knowledge base is managed as a unified resource and can be bound to a designated account, persona or digital employee to control who can reference which copy.

Media library

Images, videos, and files are stored in a centralized manner after uploading, and can be directly referenced during publishing and conversations. Supported formats and capacities

Business prompt words

Write your business rules and answer standards as prompt words as constraints when AI is generated, so you don’t have to repeat them every time in the conversation.

For AI reference

Knowledge base content is used as the priority basis for responses. Answers that hit the knowledge base are more controllable than those generated by the model itself.

Long-term memory

The needs, preferences, and progress the customer has made are retained across sessions. You can take over by changing your account or chatting on another day.

Historical conversations and conversations

Session records are retained by customer and time and can be retrieved. Used for review, handover and troubleshooting of incorrect AI answers.

Warehouse and export

Customer and conversation data are entered into the data warehouse and can be reviewed by dimension. Customer files can be exported for storage or delivery to other systems.

How it works

From uploading data to using AI

You need to confirm the two steps of warehousing and caliber. Once the knowledge base takes effect, AI will treat the content inside as facts and speak to the outside world, and wrong content will be amplified.

01
Organize and upload information

Product information, FAQs, and quotations are entered into the knowledge base; pictures, videos, and documents are entered into the material library. The library can be divided into product lines or markets.

Automatic
02
Confirm external caliber

You should review the quotation range, commitment boundaries, and compliance statements before putting them into storage. Different markets have different calibers and can be maintained separately in separate databases.

Needs you
03
Bind to personas and employees

Bind the knowledge base to a designated account, persona or digital employee. The same account references the same information in different tasks.

Automatic
04
Refer to and accumulate memories during conversations

When AI replies, it gives priority to fetching knowledge base content, and at the same time writes the needs and preferences mentioned by customers into long-term memory for subsequent conversations.

Automatic
05
Spot check and fix leaks

Sampling from historical conversations to see if the AI’s answers are accurate. If you answer a question incorrectly or cannot answer it, come back to make up the knowledge base entry or change the prompt word.

Needs you

This step of spot checks cannot be omitted. The knowledge base is AI’s source of truth. An expired quotation or a crooked promise will be told consistently to all customers who ask - errors are magnified here, not averaged. Therefore, you need to review the items before they are put into the database, and you need to sample them regularly after they go online.

Prerequisites

Specifications and prerequisites

The knowledge base is a container, and you have to provide the content. Write down clearly what you must prepare yourself and what you can make up on the go.

your business information
Product description, parameters, FAQ, quotation. The platform provides storage and retrieval capabilities and does not write content on your behalf.
Judgment of external caliber
It is up to you to confirm what can and cannot be promised and the compliance requirements of different markets. The platform does not determine legal boundaries for you.
Maintenance person
The knowledge base needs to be updated by someone. If the data has expired and no one has changed it, the AI ​​will always answer according to the old content.
storage capacitySized to you
The number of knowledge base entries and material library capacity are activated according to your actual usage, with upper limit and pricing, confirmed by the consultant.
Model capabilitiesSized to you
Retrieval and generation rely on large models. Optional models and context capacity are configured according to the plan, and the performance under different configurations is different.
Business prompt wordsOptional
It can be used without writing. After writing it, the caliber generated by the AI ​​is more stable, and it is worth coming back to make up for it after running through a few rounds of dialogue.
Works with

Who it works with

FAQ

FAQ

Will AI be able to use the content in the knowledge base correctly?

No. The knowledge base narrows the AI's answers to the scope of your information, and the hit rate is much higher than letting it answer based on common sense. However, there are still several situations where problems may occur: the customer's question and the wording of the entry are far different, and the search may not be able to retrieve it; there are two inconsistent answers to the same question in the database, and it may get the older one; the question spans several pieces of information, and the meaning it puts together may not be what you want to express. Therefore, you need to review the external wording, check historical conversations regularly, and add entries when an answer is wrong. Treat the knowledge base as an asset that needs maintenance, not a switch that you configure once and forget about.

What is recorded in long-term memory, and how to deal with customer privacy?

What is recorded is the needs, preferences and advancement progress expressed by the customer during the conversation, which will be used to continue the subsequent conversation. Record your account data, which can be viewed, modified and deleted. The specific data retention period and processing terms can be found in the data processing agreement. For cross-border scenarios, it is recommended to confirm separately according to the regulations of the target market.

What file formats does the knowledge base support?

The material library supports the upload of pictures, videos and files, and the knowledge base entries are mainly text content. Specific supported format list and single file size limit, the actual activation shall be subject to the background instructions.

Different markets have different standards. How to manage them separately?

Build multiple knowledge bases, maintain them separately by product line or market, and then tie them to the corresponding accounts or personalities. In this way, the same question on accounts in different markets will be answered according to their respective libraries, and there will be no confusion.

Can historical conversations be exported?

Yes. Customer files support exporting files, and historical sessions are saved by customer and time and can be retrieved. Used for review, handover or importing into your own system.

If the knowledge base is updated, will the ongoing dialogue change accordingly?

Subsequent replies will use new content. Messages that have already been sent will not be modified retroactively. When changing important calibers, it is recommended to also check whether there are any hard-coded old sayings in the quick reply.

First build a knowledge base and ask it some real questions

The consultant will use your information to build the first knowledge base and test it on several questions that your customers often ask.