Connections

A connection tells 3sparks Chat how to reach an API endpoint. Use Connections for cloud providers, local servers on your network, and custom OpenAI-compatible or Anthropic-compatible APIs.

Do not create a connection for on-device Local Models. For those, open Local Models, download a supported GGUF model, load it, and start a chat with the loaded Local Model.

Use Connections for

• Cloud providers such as OpenAI, Anthropic, Mistral, Gemini, OpenRouter, and Azure OpenAI
• Local servers such as LM Studio, Ollama, OpenWebUI, AnythingLLM, and other reachable APIs
• Custom compatible endpoints
• The AI Keyboard, which needs a reachable cloud or server connection

Provider basics

Each connection has:

• A provider, such as OpenAI, Anthropic, LM Studio, Ollama, or Custom
• A base URL
• A route prefix when the API needs one
• An API profile, such as OpenAI, Anthropic, or LM Studio v1
• An optional API token stored in Keychain
• An optional default model

Connection names are optional. If you leave the name blank, the app previews and saves a generated provider-based name.

Add a connection

1. Open Connections.
2. Tap +.
3. Choose a provider.
4. Enter the base URL, API key, or other settings the provider needs.
5. Leave the name blank to let 3sparks Chat generate one, or enter your own.
6. Tap Create Connection.
7. Tap Fetch Models when the provider supports it.
8. Choose a default model.
9. Tap Done.

If you only want private on-device GGUF model use, skip this flow and use Local Models instead.

New chat defaults

A connection also decides how its new chats begin:

• Enable Memory for New Chats — whether chats on this connection start with memory on, or start as private chats.
• Default Instruction — the instruction selected automatically for new chats.

These apply as soon as you choose the connection as a chat target, and you can override them before creating the chat. Marking a connection Favorite also makes it the one that tapping New Chat uses. See Starting a Chat and Private Chats.

Fetch models

After saving a connection, tap Fetch Models when the provider supports it. Fetching models helps the app show the current models available to your account or server.

If a selected model later becomes unavailable, the app may show a warning in chat. Use Switch Model or fetch models again from the connection.

Token limits

Token limits help avoid sending more context than a model can handle.

• Leave token limits unset for normal use.
• Use presets for known provider limits.
• Use a custom limit for smaller local-server models or strict providers.

Output token limits are separate from context limits. Output tokens control response length. Context limits control how much prior chat and attached content can be sent.

Local servers

For LM Studio, Ollama, OpenWebUI, AnythingLLM, and similar servers, use an address your iPhone or iPad can reach:

• Same network: http://YOUR_LOCAL_IP:PORT
• Away from home: a Tailscale IP or a public HTTPS endpoint

Do not use localhost from a phone unless the server is running on the phone.

Local servers are different from on-device Local Models. A local server runs outside 3sparks Chat and exposes an API. On-device Local Models run inside the 3sparks Chat app after you download and load them from the Local Models view.

Common recovery paths

• Unreachable server: confirm the server is running and reachable from the device.
• Connected with no models: tap Fetch Models or load a model on the server.
• No selected model: pick a default model on the connection or in chat settings.
• Selected model unavailable: use Switch Model or fetch models again.
• API token rejected: create a new provider token and update the connection.
• Looking for on-device models: open Local Models instead of Connections.