Skip to main content

FAQs

1. Which models does LinkAI support?​

General LLM Models: ChatGPT series (OpenAI), Claude series (Anthropic), Gemini series (Google), DeepSeek series, Qwen series (Alibaba);

Text-to-Image Models: Dall-E-3 model

Image Recognition Models: OpenAI image recognition model, Claude image recognition model, OCR model;

Voice Models: Speech recognition: OpenAI Whisper; Speech synthesis: OpenAI

2. What if responses don't meet expectations?​

Such as not following Agent settings, not retrieving from knowledge base, or not executing workflows as expected

If you find that the AI responses don't meet expectations when using LinkAI (especially through integration channels), first check the record details in the Console - Account Page under Usage Records. The usage record details will show comprehensive information about the conversation, including the Agent used, workflow, model, plugins, knowledge base hits, conversation context, etc. For workflows, you can also view the execution process and the input/output at each step. Self-checking these records can help identify or solve 90% of issues.

3. What is the principle and function of LinkAI's knowledge base?​

LinkAI's knowledge base feature is built on a self-developed RAG (Retrieval-Augmented Generation) model. It pre-processes long texts by splitting and vectorizing them for storage. When a user asks a question, it retrieves relevant content through vector search, then leverages the language model's semantic understanding and generation capabilities to formulate a response. We continuously optimize the knowledge base's accuracy, performance, and capacity based on different application scenarios.

Knowledge base applications provide customized intelligent Q&A services based on uploaded text data, which helps address the issues of general language models lacking domain-specific data support and occasionally producing "hallucinations." With a knowledge base, you can create intelligent customer service based on product manuals or FAQs, interact with learning materials to study specific knowledge, or even upload a novel to create a character persona.

4. How should I choose between "Unstructured Documents, QA Format, Tables, and Website Import" for knowledge base?​

The choice of knowledge base import method depends on your existing knowledge resources and application scenarios. You can combine multiple file import methods to achieve better results. For example, in customer service scenarios, you can organize customer service FAQs into Q&A format to ensure accurate answers to high-frequency questions. For product information or service description tables with multiple columns, you can import them as tables to provide precise information retrieval capabilities. For more open-ended conversation scenarios, you can import company introductions, product descriptions, and usage guides as unstructured documents. Personal blogs, company websites, and other online resources can be imported through website parsing to cover long-tail questions and reduce the no-result rate.

5. What file formats does the knowledge base support?​

  • Document types: Currently supports unstructured files in txt, pdf, md, docx formats

  • Q&A types: Currently supports QA (question and answer) format in csv (two-column) files

  • Table types: Currently supports csv (multi-column) files and Excel (multi-column) files

  • Website import: Currently supports submitting direct webpage URLs or sitemap URLs to automatically parse webpage content and import it into the knowledge base.

6. What is the principle and function of the database feature?​

LinkAI's database feature allows AI to access structured data by creating built-in databases or connecting to remote (proprietary) databases. Through AI's understanding of tables and fields in the database, it can generate SQL queries based on user questions to query the database.

Agents or workflows connected to databases can perform conversational data analysis, automatically generating SQL statements from natural language, querying data, analyzing results, and generating data charts, all supported across web interfaces and various integration channels. They can also perform dynamic data management, such as recording customer feedback, order records, or survey results during conversations with users.

7. When importing knowledge data, how should I choose between database and knowledge base?​

Knowledge base uses RAG (Retrieval-Augmented Generation) technology, leveraging semantic vector matching, keyword matching, and other methods to search through massive unstructured data (such as unstructured Word document content). It then uses large language models to analyze the retrieved results and user questions to generate answers.

Database, on the other hand, is a container for structured data (database-table-field-field value). When using AI to access databases, the core capability is the AI's ability to autonomously generate data query code (SQL) and analyze the queried data.

Therefore, knowledge base retrieval is suitable for unstructured data, scenarios with lower data integrity requirements, and more qualitative answers. Database queries are suitable for structured data, scenarios with high data integrity requirements, and statistical analysis that requires more quantitative answers.

8. What if responses from custom integration clients don't match the knowledge base or aren't accurate enough?​

After creating a knowledge base Agent on the web interface, you can click "Start Conversation" to ask the same question and check if the response matches what you get from the client. If there's a significant difference, there might be an issue with the client configuration. Verify that you've correctly set the Agent code (note that this is not the knowledge base code; the knowledge base needs to be bound to an Agent to work).

9. What if web interface conversations don't seem to match the knowledge base or aren't accurate enough?​

You can optimize knowledge base retrieval and response quality by using the Search Test feature and adjusting relevant settings in the Agent (similarity threshold, number of retrievals per query, no-hit strategy). If you're currently using semantic retrieval, you can try enhanced retrieval, which combines semantic retrieval with full-text keyword search for better precision with names, letters, numbers, and codes. See the Knowledge Base Troubleshooting documentation for more information.

10. How can I quickly integrate with WhatsApp, Discord, Slack, LINE, Telegram, HubSpot, and other social or office software?​

You can use LinkAI's Pro/Team version one-click integration feature. For detailed integration documentation, refer to Channel Integration.

11. Can I integrate LinkAI into my own applications?​

Yes. You can quickly integrate LinkAI into your applications using our API. Refer to the API Documentation.

12. Can I use LinkAI's key in other OpenAI projects?​

Yes. Simply refer to the API documentation, and where you would normally enter open_ai_api_key, enter your linkai api key instead. For the openai.api_base or custom proxy address, enter https://api.linkai.cloud/v1. Note that some projects may not require the /v1 suffix.

13. How do I change my password? What if I forget my password?​

You can log in by scanning the WeChat QR code, then click on your avatar in the upper right corner, select "My Account," and click the Edit button next to Personal Information to change your login password. If you haven't linked a WeChat account, you can click "Forgot Password" directly in the login window.

14. How does LinkAI charge?​

LinkAI platform offers three subscription tiers based on resource capacity, feature benefits, and service support: Free, Pro, and Team versions, providing options for customers with different needs. For specific version details, please refer to the Version Description. Model services use a credit system, charging based on tokens consumed during conversations. For detailed pricing information, please refer to the Pricing Rules.