Connect providers and local models
Open Settings → AI Services. Choose a provider for the AI tasks you need, including Agent work, text, speech, images, search, and extraction. Depending on the service, connect with an API key, a supported sign-in method, or a local server address.
You can connect more than one provider. Supported Ollama and LM Studio servers provide a local-model option. A connection makes supported models available; it does not guarantee that every model supports every task.
Check a connection
Use Test Connection to check that a service can respond. If it fails, review the displayed error, credentials, network connection, and service availability. For a local model, check that the server is running and the intended model is available.
A successful check applies to that test. A later request can still fail because of rate limits, provider errors, unsupported input, or a lost connection.
See available AI capabilities
Open Settings → AI Capabilities. A capability describes a task rather than a particular model.
- Agent
- Multi-step work with a Builder or Runtime Agent.
- Text and image understanding
- Generate text or analyze supported image input.
- Decision Making
- Use TypeSafe Jev to judge an input against stated criteria, choose a category, or return a score for an app workflow.
- Text Embedding
- Use supported OpenAI or Google models to represent text for meaning-based search, matching, or retrieval.
- Image generation
- Create or edit images through a supported service.
- Speech to text and text to speech
- Transcribe audio or produce spoken audio.
- Music and video generation
- Request media from a configured service that supports it.
- Web Search and Web Extract
- Search for information, or extract readable content from an eligible public web page.
The settings show ready models and supported usage paths. For a provider that needs setup, choose Configure to open it in AI Services. This includes Web Extract services such as Firecrawl. Services that need no sign-in have no Configure button.
Web Extract reads eligible public pages. It does not read private intranet pages, local files, or every online document URL.
Choose models and their order
Each capability has an ordered list of models. Drag ready models into your preferred order. AppJuice uses this preference when an app or operation does not choose a specific model.
Use the conversation's model picker to choose a model and supported Thinking level for subsequent work. You can change them without starting another conversation. The capability order does not override the Builder picker. Choose All available models to see more than the recommended choices.
Use AI when building and using apps
- Builder
- Uses a capability while creating or changing an app.
- Runtime Agent
- Uses a capability as a tool while carrying out a task inside the app.
- Direct call
- The app calls a shared AI helper for a defined action, such as transcribing an upload.
These features can use the AI services you have already connected in AppJuice. To add one to your app, describe the result you want in Tweak. See Add AI features to your app for example requests.
Choose speech services and voices
Configure Speech to Text for dictation. The microphone tooltip shows the selected service and model, or explains why dictation is unavailable.
Configure Text to Speech for spoken output. Ask Builder to add a voice or language selector if needed. Available voices depend on the service; supported catalogs include Google, MiniMax, Soniox, and Edge TTS.
Check which service answered
Replies show the provider and model that answered. Supported AI tool activity, including Web Search, also shows its service and model. An Agent may use a different service for an image or audio task.
Understand usage and cost
Cloud services apply their own pricing and account limits. AppJuice can show quota information where a connected service exposes it. Missing quota information does not mean unlimited use or zero cost.
Check your provider account for authoritative billing. Local models use the host computer's resources. Online capabilities still use their configured services even when the main Agent model runs locally.