Model Provider Support
Open Notebook supports multiple AI model providers to give you flexibility in choosing the AI that best fits your needs.
Provider | Highlights |
---|---|
OpenAI | Great models, covering all necessary features for Open Notebook |
Anthropic | Very capable Sonnet 3.5 for dynamic reasoning |
Gemini | Large context (2M tokens) and the best text to speech for podcasts |
Ollama | Running local models for free. Great for transformation tasks |
ElevenLabs | For amazing voice quality |
Open Router | Great option for using several open source models, as well as Cohere, Mistral, xAI, etc |
Groq | Very fast inference, but limited model availability |
xAI | The powerful grok model, less guardrails, great responses. |
Vertex AI | If you are running a Google Cloud environment |
All providers are installed out of the box. All you need to do is to setup the environment variable configurations (API Keys, etc) for your selected provider and decide which models to use.
Please refer to the .env.example
file for instructions on which ENV variables are necessary for each.
API Key Requirements
The Podcast Generator feature currently requires Gemini API keys for content generation. Additionally, voice generation requires either an OpenAI API key or an Eleven Labs API key. Make sure you have the necessary API keys configured before using these features.
Create models on the Settings page
Go to the settings page and create your different models.
📝 Notice: For complete usage of all the features, you need to setup at least 4 models (one of each type).
Model Type | Supported Providers |
---|---|
Language | OpenAI, Anthropic, Open Router, LiteLLM, Vertex AI, Vertex AI, Anthropic, Gemini, Ollama, xAI, Groq |
Embedding | OpenAI, Gemini, Vertex AI, Ollama |
Speech to Text | OpenAI, Groq |
Text to Speech | OpenAI, ElevenLabs, Gemini |
If you are not sure which models to setup, the Model Settings page will offer some options for you to get started with.
After setting up the models, head to the Model Defaults tab to define the default models. There are several defaults to setup.
Model Default | Purpose |
---|---|
Chat Model | Will be used on all chats |
Transformation Model | Will be used for summaries, insights, etc |
Large Context | For content higher then 110k tokens (use Gemini here) |
Speech to Text | For transcribing text from your audio/video uploads |
Text to Speech | For generating podcasts |
Embedding | For creating vector representation of content |
All model types and defaults are required for now. If you are not sure which to pick, go with OpenAI, the only one that covers all possible model types.
The reason for opting for this route is because different LLMs, will behave better/worse depending on the type of request and type of tools offered. So it makes sense to build a more refined system to decide which model should process which task.
For instance, we can use an Ollama based model, like gemma2
to do summarization and document query, and use openai/claude for the chat. The whole idea is to allow you to experiment on cost/performance.
Suggested Configurations
These are some suggested configurations for different use cases and budgets:
Best in Class
Model Default | Model Name |
---|---|
Chat Model | claude-3-5-sonnet-latest |
Transformation Model | gpt-4o-mini |
Large Context | gemini-1.5-pro |
Speech to Text | whisper-1 |
Text to Speech | eleven_turbo_v2_5 (elevenlabs) |
Embedding | text-embedding-3-small |
Open AI Only Configuration
Model Default | Model Name |
---|---|
Chat Model | gpt-4o-mini |
Transformation Model | gpt-4o-mini |
Large Context | gpt-4o-mini (you will be limited to 128k tokens) |
Speech to Text | whisper-1 |
Text to Speech | tts-1-hd |
Embedding | text-embedding-3-small |
Gemini Only Configuration
Model Default | Model Name |
---|---|
Chat Model | gemini-1.5-flash |
Transformation Model | gemini-1.5-flash |
Large Context | gemini-1.5-pro |
Speech to Text | (not available yet) |
Text to Speech | default |
Embedding | text-embedding-004 |
Open Source Only (using Ollama)
Model Default | Model Name |
---|---|
Chat Model | qwen2.5 or gemma2 or phi3 or llama3.2 |
Transformation Model | qwen2.5 or gemma2 or phi3 or llama3.2 |
Large Context | qwen2.5 or gemma2 or phi3 or llama3.2 (limited to 128k) |
Speech to Text | (not possible yet) |
Text to Speech | (not possible yet) |
Embedding | mxbai-embed-large |
We are working hard to support more providers and model types to give users more flexibility and options.
Testing your models
If you are not sure which model will work best for you, you can try them up on the Playground section and see for yourself how they handle different tasks.
⚠️ Important instructions for Gemini
The new Gemini Text to Speech models are amazing and definitely worth using. But in order to use them, you need to do a little setup. Please refer to this Podcastfy help page for details. But it basically requires you to enable the Text to Speech API and add it to your API Key.