Configure AI services
Configure text models, transcription, and optional web search.
Which features need configuration
| Service | Purpose | Required? |
|---|---|---|
| OpenAI compatible text model | Chat, summaries, knowledge points, exercises, review materials, and other AI tasks | Required for these features |
| Tencent Cloud speech recognition | Turn recordings and imported audio into text | Required for transcription |
| Tavily | Web search in conversations | Optional |
Your chosen providers supply these services and may charge for requests. Saving and finding local resources does not require configuring every service.
Configure a text model
- Open “设置 → AI 服务” (Settings → AI services).
- Enter the provider's service URL, model name, and API key in the AI service section.
- Save, then use “测试连接” (Test connection) if needed.
- After a successful test, return to the record or conversation and retry the AI task.
You can use “填入 DashScope 预设” (Fill DashScope preset) as a starting point, then check the details against your account. Only one text model configuration is active at a time; assistants do not each have an independently configured model.
Connection testing sends a real model request and may incur a charge. Saving keys does not authorize sending resources. Content sharing consent is handled separately.
Configure transcription
In “语音转写” (Transcription) on the same page, enter Tencent Cloud SecretId, SecretKey, and region. Save and test if needed.
Before the first transcription, consent is required to send audio to Tencent Cloud. If keys or consent are missing, check the waiting reason in the record and complete the requirements. Do not repeatedly import the same file.
Configure web search
Save your Tavily key in “联网搜索” (Web search), then enable web search in a general assistant conversation. Without configuration, the switch is unavailable; other AI features remain usable.
Search terms are sent to Tavily, and connection testing also runs a real search. The local knowledge assistant does not provide web search.
When configuration fails
- Connection failed: check the provider's compatible endpoint, model availability, network, and key permissions.
- Insufficient credits or rate limiting: review the provider account and error message, resolve the issue, then retry.
- Recording keeps waiting: check transcription keys, network, and consent to send audio.
Do not include real keys in public screenshots, issues, or docs. See Data and privacy for data and sharing scope.