By use case

By industry

Let customers ask about products, check orders and get support without leaving your store. You continue to own the service entry point, brand and customer relationship.

Who this is for

This solution is for merchants with a website, customer accounts and an order system, and developers building stores for them. Prepare your product information, approved support policies and human contact details. Your business system must be able to look up orders for the authenticated customer.

Start with one store and one task, such as order lookup for signed-in customers. Clear answer sources and escalation conditions make the first rollout easier to verify. The demo accounts and orders are examples, not a customer system you can use for your business.

What customers can do

• Ask while shopping. Customers open a support bubble on the store page. You choose its name, greeting and the questions it should answer.
• Check their own orders. Once you connect an order tool, the assistant can read the current customer's order status without asking them to repeat every detail.
• Request a return. Check the order and eligibility before asking the customer to confirm. The demo submits a request; it does not promise automatic refunds or a fixed settlement time.

These actions depend on the business tools you connect. The tool's server must verify order ownership. Changing a name in a conversation must not reveal another customer's orders. Policies absent from your material should lead to your support contact rather than a guessed answer.

Solution components

Launch the solution

1. Follow the agent quickstart. Add approved support policies and check the answers in a trial conversation.
2. Follow web embedding to connect the store. Customer identity comes from your website's authenticated session; credentials stay on your backend.
3. Use tool authorization to connect order lookup, then verify return requests using test accounts and orders.
4. Check two accounts for order isolation, sign-out behavior, unknown policy questions and request confirmation. Confirm the business record actually changes before opening support to customers. Keep the original human support entry point available.

Give this to your agent:

Connect support to my store. Read /en/docs/solutions/ecommerce-support on this site and follow its integration links.
Confirm my store URL, test environment, agent, login method, order interfaces and approved policies. Configure credentials only on the backend; never print them in the conversation.
Done means: chat works; two test accounts can only see their own orders; a confirmed return request reaches the business system; unknown policies lead to human support.
Deliver the integration, verification records and instructions for disabling the support entry point.

Cost estimate

Cost estimate pending. Public pricing and the model catalog do not establish the full cost of this solution. During the trial, record conversation counts, turns per conversation, model usage and tool calls before calculating a budget from verifiable prices. No fixed cost per order is promised here.

Demo

Open the Good Finds support demo, select a demo account in the top-right corner and open the support bubble. Ask “查看我的订单” (“Show my orders”) and compare the response with the orders page. Use demo orders for return requests and do not enter real customer data.

Next step

Complete the web embedding guide, then inspect conversations and usage through operations. If you only need answers about public policies, begin with the help center solution.

Give learners explanations and revision prompts relevant to the lesson they are watching, without leaving your course. You retain ownership of the content, brand and learner relationship.

Who this is for

This solution is for education providers, course creators and developers operating a learning website. Prepare lesson descriptions, key points and answer boundaries. You need to be able to modify the course page and connect its existing login system.

Start with one course whose material is complete. The assistant answers from the content and page state you provide; a video on the page does not mean it automatically understands the recording. Questions outside the course, exam scoring and individual learning judgments should have a clear route to a teacher.

What learners can do

• Ask about the current lesson. Pass the course and lesson information to the assistant so learners do not have to explain where they are each time.
• Check playback progress. A page tool can read the player's position, giving questions such as “Where did I get to?” a concrete answer.
• Adjust playback from the conversation. With the corresponding page tools connected, the assistant can act on requests to seek to a position or switch lessons. Your player implements each action.

Page information helps explain the learning context. It does not determine whether a learner purchased a course. Your website remains responsible for course access, and browser-reported lesson names or progress are not evidence of grades or completion.

Solution components

Launch the solution

1. Follow the agent quickstart. Prepare the course description, lesson notes and a response policy for questions outside the material.
2. Follow web embedding to place the assistant in the course page. Your backend issues tokens for the signed-in learner.
3. Follow client tools and page context to connect the lesson and player. Start by reading progress, then add seeking or lesson switching where needed.
4. Ask questions at different lessons and playback positions, checking that answers use the current state. Test account changes, sign-out and tool failures. Have a teacher approve the answer scope before opening the assistant to learners.

Give this to your agent:

Add an assistant to my course website. Read /en/docs/solutions/course-assistant on this site and follow its integration links.
Confirm the course URL, login method, agent, lesson materials, player interface and allowed playback actions. Configure credentials only on the backend.
Done means: chat works; switching lessons updates the assistant's context; reported progress matches the player; allowed playback actions can be verified; questions outside the course lead to a teacher.
Deliver verification records and a way to disable the assistant without disrupting course playback.

Cost estimate

Cost estimate pending. Public pricing does not establish the complete cost of a course assistant. Trial one course and measure active learners, conversation turns per learner and model usage before budgeting. Track video hosting and existing learning-system costs separately.

Demo

Open the Good Finds course assistant, choose a demo account and open the assistant. Switch lessons, move the playback position and ask “我现在学到哪里了？” (“Where am I in the course?”). Compare the answer with the page.

The demo uses a simulated player to show interaction between lesson state, playback progress and the assistant. It does not analyze a real course video. You can also request a different playback position and observe whether the player performs the action.

Next step

Use the client tools guide to connect your own player. Before launch, read embedding security and keep a teacher contact option available for questions the assistant cannot answer.

Let customers ask questions in your help center and find answers in your published material. You maintain the help content, brand and customer contact channels.

Who this is for

This solution is for businesses with product documentation, frequently asked questions or support policies, and developers maintaining those sites. Prepare reviewed material that can be made public, a page you can modify and a backend that can establish visitor identity.

Start with a clearly defined subject, such as delivery, returns or membership. A small collection can go directly into the assistant's instructions; evaluate knowledge retrieval separately when the collection grows. This demo answers from a fixed FAQ. It does not promise to crawl your site or keep material up to date automatically.

What customers can do

• Ask on the original page. Customers click a search entry point or use the page's keyboard shortcut to open a conversation without moving to another chat product.
• Get answers from published content. Turn frequently asked questions into a conversation where customers can follow up on the same topic.
• Find help beyond the material. If discounts, timings or service commitments are absent from the source, direct customers to your support contact for confirmation rather than guessing.

This assistant explains help content. Knowing the return policy does not give it permission to read orders or issue refunds. If you later need account-specific data, connect authorized tools for those business actions and verify them separately.

Solution components

Launch the solution

1. Follow the agent quickstart. Prepare public material and a support contact. Check trial answers against the source before adding the assistant to your site.
2. Follow the docs Q&A recipe for the window and entry point. Use existing login identity where available. For visitors, have your backend establish a stable, separate identity for each person.
3. Follow embedding security to check allowed website origins and backend credential storage. The demo's shared guest identity is for demonstration; do not copy it as the identity design for your live site.
4. Test known and unknown questions, closing and reopening the window, different visitors' conversations and loading failures. Keep the original help articles and support contact available so customers can still get help if chat is unavailable.

Give this to your agent:

Add Q&A to my help center. Read /en/docs/solutions/help-center-assistant on this site and follow its integration links.
Confirm the help-center URL, approved public material, agent, login or visitor identity design and human support contact. Configure credentials only on the backend.
Done means: the entry point opens working chat; known answers match the source; unknown answers do not invent policy; visitors' conversations are isolated; failures leave articles and human support accessible.
Deliver verification records and instructions for disabling the Q&A entry point.

Cost estimate

Cost estimate pending. Public pricing and the model catalog alone do not establish the cost of the complete service. Record question volume, average turns, source length and actual model usage during the trial. If you later add knowledge retrieval, account for its costs separately.

Demo

Open the Good Finds help center. Click the search entry or press ⌘K, or Ctrl+K on Windows and Linux. Ask “如何申请退货？” (“How do I request a return?”), then ask for a specific refund time absent from the material.

Compare answers with the FAQ on the page: known rules should match, while missing commitments should lead to customer support. The example uses store policies; replace them with your own help content when integrating your product.

Next step

Start with the docs Q&A recipe. For customer order lookup or return requests, continue with the ecommerce support solution and add authorization and verification for each new action.