Custom Provider

Connect BraceKit to any AI API endpoint with support for multiple formats: OpenAI, Anthropic, Gemini, and Ollama. Perfect for self-hosted models, proxies, or alternative services.

Use Cases

• Local models: LM Studio, Jan.ai, LocalAI
• Proxies: OpenRouter, Azure OpenAI
• Self-hosted: vLLM, TGI, text-generation-webui
• Alternative services: Any OpenAI-compatible API

Setup

1. Open Settings

1. Click Settings (⚙️)
2. Go to AI Provider tab

2. Add Custom Provider

Click the + Add button next to the provider dropdown and fill in:

Click Save Provider to create it.

3. Configure and Use

1. The new provider appears at the top of the provider dropdown — select it to activate
2. Enter your API Key in the Configuration section (or leave empty for local servers)
3. Click Add model and enter the model id the API expects (for example llama-3.1-8b). Fill in specs you know:
   • Context window and max output (so compact and the token meter stay accurate)
   • Input modalities and Vision if the model can read images
   • Tools and Reasoning if the model supports them
4. Save. You can edit a model later with the pencil icon, or change the same fields on the Advanced card.

If BraceKit can list models from your endpoint, new names are added to the list. Specs you already saved are never overwritten.

API Formats

BraceKit supports multiple API formats:

OpenAI Format

The most common format, used by:

• LM Studio
• Jan.ai
• LocalAI
• vLLM
• OpenRouter
• Azure OpenAI

Base URL: http://localhost:1234/v1
Endpoint: /chat/completions

Anthropic Format

For Anthropic-compatible endpoints:

Base URL: https://your-anthropic-proxy.com
Format: Anthropic

Gemini Format

For Gemini-compatible endpoints:

Base URL: https://your-gemini-proxy.com
Format: Gemini

Ollama Format

For Ollama native API:

Base URL: http://localhost:11434
Format: Ollama
Endpoint: /api/chat

Common Configurations

LM Studio

1. Open LM Studio
2. Start a local server (port 1234)
3. Configure in BraceKit:

Name: LM Studio
Base URL: http://localhost:1234/v1
API Key: none
Format: OpenAI

Jan.ai

1. Open Jan
2. Enable local server in settings
3. Configure in BraceKit:

Name: Jan
Base URL: http://localhost:1337/v1
API Key: none
Format: OpenAI

LocalAI

Name: LocalAI
Base URL: http://localhost:8080/v1
API Key: none
Format: OpenAI

vLLM

Name: vLLM
Base URL: http://localhost:8000/v1
API Key: none
Format: OpenAI

OpenRouter

Name: OpenRouter
Base URL: https://openrouter.ai/api/v1
API Key: your-openrouter-key
Format: OpenAI

Azure OpenAI

Name: Azure OpenAI
Base URL: https://your-resource.openai.azure.com/openai/deployments/your-deployment
API Key: your-azure-key
Format: OpenAI

Model Fetching

BraceKit attempts to fetch available models from the /models endpoint.

If Fetching Works

Models appear automatically in the dropdown.

If Fetching Fails

1. Type the model name manually
2. Check the API documentation for available models
3. Verify the endpoint is correct

Troubleshooting

“Connection refused”

• Ensure the server is running
• Check the port is correct
• Verify no firewall is blocking

“401 Unauthorized”

• Add the correct API key
• Some services need any non-empty string

“Model not found”

• Type the model name manually
• Check the service’s model list
• Verify model name spelling

“CORS error”

• The server may need CORS headers
• Configure the server to allow browser requests
• Or use a browser extension to bypass (development only)

Streaming not working

• Ensure the endpoint supports SSE
• Check the response format matches the selected format
• Some local servers don’t support streaming

Security Notes

Local Servers

For local development servers:

• No API key needed (or any placeholder)
• Only accessible from your machine
• Safe to use without authentication

Remote Servers

For remote or public servers:

• Always use a real API key
• Ensure HTTPS is enabled
• Consider rate limiting

Related

• Ollama — Recommended for local models
• Configuration — All settings