Connect a model

The model service interprets and generates content. Lexora brings conversations, files, tools, and execution into the same workspace. Before starting, authenticate a service and enable the models you want to use.

Choose a connection type

Open Settings → Model services → Add service.

What you have	Choose	Prepare
A listed provider	Built-in services	An API key or an OAuth account supported by that service
A compatible API or company gateway	Custom API	API type, endpoint, credentials, and model ID
A local or LAN model server	Custom API	A running compatible endpoint, model ID, and any required credentials

Lexora does not install or start local model servers. Start the service according to its own instructions, then add its connection.

Add a built-in service

1. Find the provider under Built-in and choose Add.
2. In Configure connection, choose Configure API key or Use OAuth. Obtain keys from the provider's console; follow the browser instructions for OAuth.
3. In the model configuration step, enable the models you need.
4. Choose Finish and enable. Return to a task, select the model, and send a short message to check the connection.

Enter credentials through the dedicated setup dialog, not in a task conversation.View original

Enter credentials through the dedicated setup dialog, not in a task conversation.

A saved-key status confirms storage. Successful requests still depend on valid credentials, model access, balance, and networking. A short conversation is the most direct check.

Add a custom API

Under Custom API, fill in the display name, identifier, API type, and endpoint, then configure authentication.

• Display name helps you recognize the connection, such as “Home model server.”
• Identifier distinguishes configurations. Follow the form's rules and avoid duplicates.
• API type must match the protocol the server actually provides. Supporting one interface does not imply support for all interfaces.
• Endpoint is the API base URL from the service documentation, not its chat website.
• Model ID is the name the server accepts, not just a display label you invented.

After authentication, use Sync models if discovery is supported. Otherwise, choose Add model manually. In either case, explicitly enable the models you want to use.

When should I change advanced settings?

Only when the service needs extra request headers or you deliberately need to override model parameters. Context window, output limits, and input capabilities should match the service. Raising a number does not raise its server-side limit. Get a minimal request working with defaults first, then change one setting at a time.

Connect a local AI model

If you run a model service on your computer or local network, connect its compatible API to Lexora:

1. Start the service and check that it has loaded a model and exposed its API.
2. Add a Custom API with the protocol, base URL, and credentials it requires. localhost means the computer running Lexora; a service on another machine needs an address that Lexora can reach.
3. Discover or manually add a model, enable it, and send a short message to check the connection.

The local model handles inference; Lexora organizes tasks and tool execution. A local model does not make every action offline: web tools, remote MCP servers, and backups can still use the network.

Choose models for tasks and companions

Model selection is associated with companions. Set a default model in a companion's configuration, or select one for that companion in a task. Choosing a DeepSeek, Claude, or ChatGPT avatar does not connect the corresponding provider automatically.

A fast model may suit simple organization. Images, complex material, and tool use need appropriate model capabilities. Check the actual configuration and service documentation rather than relying on the name.

Inspect usage and resolve connection problems

Settings → Usage shows the usage Lexora records, including task and model breakdowns. Your provider's console remains authoritative for billing and quotas.

Symptom	Check first
No models in the task	Service setup is complete and at least one available model is enabled
Authentication required	Key validity or whether OAuth needs authorization again
Model not found	Exact model ID and your account's access
Connection times out	Server status, endpoint, network, and proxy settings
Text works but images do not	Image support in this model and API protocol

Once connected, return to your first task, or read Companions and collaboration to assign different models to different roles.