BibGenie DocsCustom Models
Add and manage custom models, including OpenAI, Anthropic, OpenRouter, Ollama, LM Studio, and compatible APIs.
Custom models let you use your own API key, third-party provider, local model, or institutional model service. Once configured, BibGenie sends requests directly from the local Zotero plugin to the provider you choose.
Add a Model
Open BibGenie and go to Model Settings.
Click Add Model in the Custom Models section.
Select or enter a Provider ID. BibGenie loads common providers from a model catalog and automatically fills Provider Name, Provider Type, and Base URL when possible.
Select or enter a Model ID. If the model exists in the catalog, BibGenie automatically fills Model Name, Family, Model Types, Context Window, Max Tokens, and Description.
Enter your API Key. For local services such as Ollama, an API key is usually not required.
Click Test to verify the connection. A successful test shows Connected. If the test fails, BibGenie shows the provider error message so you can correct the configuration.
Review the auto-filled fields and adjust Model Types, Context Window, or Max Tokens if needed.
Click Add Model. The model is saved, enabled, and becomes available in the chat model selector.
Auto-Fill Behavior
Adding a custom model is not entirely manual. BibGenie loads provider and model information from the models.dev catalog when available:
- Provider ID supports searching common providers and entering a custom provider manually.
- Selecting a provider can auto-fill Provider Name, Provider Type, and Base URL.
- Model ID is filtered by the selected provider and supports both catalog models and custom model IDs.
- Selecting a catalog model can auto-fill Model Name, Family, Model Types, Context Window, Max Tokens, Description, and pricing metadata.
- If the model catalog cannot be loaded, BibGenie uses a built-in fallback provider list, and you can still add models manually.
Prefer catalog entries when available
If your provider and model appear in the search dropdowns, select them instead of typing everything manually. This reduces mistakes in Provider Type, Base URL, model capabilities, and context limits.
Fields
| Field | Required | Description |
|---|---|---|
| Provider ID | Required | Unique provider identifier, such as openai, anthropic, or openrouter; custom IDs are supported |
| Provider Name | Required | Display name for the provider; usually auto-filled from the catalog |
| Model ID | Required | Actual model ID used by the provider API, such as gpt-4.1 or claude-sonnet-4-5 |
| Model Name | Required | Display name shown in BibGenie; usually auto-filled from the catalog |
| API Key | Optional | Your provider API key; local providers such as Ollama usually do not require one |
| Base URL | Optional | API endpoint; often auto-filled for catalog providers, but compatible or local services may require manual confirmation |
| Provider Type | Required | API integration type, such as OpenAI, Anthropic, Ollama, or OpenAI Compatible |
| Family | Optional | Model family, such as qwen, gpt, or claude; often auto-filled |
| Model Types | Required | Capabilities such as Text, Vision, Reasoning, Function Calling, or Web Search |
| Context Window | Optional | Model context window; often auto-filled for catalog models |
| Max Tokens | Optional | Maximum output tokens; often auto-filled for catalog models |
| Description | Optional | Model description; often auto-filled |
Provider Type Guide
| Provider Type | Use when |
|---|---|
| OpenAI | You use the official OpenAI API |
| Anthropic | You use the Anthropic Claude API |
| xAI | You use the xAI Grok API |
| DeepSeek | You use the official DeepSeek API |
| You use the Gemini API | |
| Groq | You use the Groq API |
| Moonshot AI | You use Kimi / Moonshot models |
| OpenRouter | You use the OpenRouter API |
| Alibaba | Usually via OpenAI Compatible |
| Z.AI / Zhipu AI | Usually via OpenAI Compatible |
| Ollama | You use a local Ollama service |
| OpenAI Compatible | You use LM Studio, an API proxy, or an institutional compatible endpoint |
Base URL
Many official providers have a default Base URL. You usually need to edit Base URL only when:
- You use OpenAI Compatible.
- You use an API proxy.
- You use a local model service.
- Your provider requires a custom endpoint.
Examples:
https://api.openai.com/v1
https://openrouter.ai/api/v1
http://localhost:11434
http://localhost:1234/v1Watch the /v1 path
Many compatible APIs require a Base URL ending in /v1. Missing the version path can cause NetworkError, 404, or model-not-found errors.
Ollama Example
Make sure Ollama is installed and running.
Run ollama list in a terminal to confirm a local model is available.
If Ollama is not yet running, start it with ollama serve. You should see output confirming it is listening on 127.0.0.1:11434.
Add a model in BibGenie and choose Ollama as Provider Type.
Use http://localhost:11434 as Base URL.
Use the model name shown by ollama list, such as llama3.1:8b, as Model ID.
LM Studio Example
LM Studio is usually configured through OpenAI Compatible:
Start the local API server in LM Studio.
Add a model in BibGenie and choose OpenAI Compatible as Provider Type.
Enter the LM Studio API address as Base URL, usually similar to http://localhost:1234/v1.
Enter the ID of the model currently loaded in LM Studio as Model ID.
Model Types
Select capabilities according to what the model actually supports:
- Text: text chat; usually required.
- Vision: image, screenshot, and figure input.
- Reasoning: useful for complex reasoning tasks.
- Function Calling: affects search, reading, and library-management workflows.
- Web Search: enable only if the provider and model support it.
Incorrect capability settings can cause ModelNotSupported or tool-calling failures.