Skip to main content
NoteWise routes every LLM call through LiteLLM, which provides a unified interface across providers. Any model string that LiteLLM supports will work with NoteWise — the table below lists the providers with native first-party support.

​
Supported providers

ProviderAPI key env varExample model stringNotes
Google GeminiGEMINI_API_KEYgemini/gemini-2.5-flashDefault provider. Free tier available. Also accepts vertex/ prefix for Vertex AI.
OpenAIOPENAI_API_KEYgpt-4oSupports gpt-4o, gpt-4o-mini, o1, o3-mini.
AnthropicANTHROPIC_API_KEYclaude-3-5-sonnet-20241022Supports Claude 3.5 Sonnet and Haiku.
GroqGROQ_API_KEYgroq/llama3-70b-8192Very fast inference. Requires groq/ prefix.
xAIXAI_API_KEYxai/grok-2Requires xai/ prefix.
MistralMISTRAL_API_KEYmistral/mistral-large-latestRequires mistral/ prefix.
CohereCOHERE_API_KEYcommand-r-plusModel name alone (no prefix) is sufficient.
DeepSeekDEEPSEEK_API_KEYdeepseek/deepseek-chatRequires deepseek/ prefix.

​
Model string format

LiteLLM model strings follow the pattern provider/model-name. NoteWise uses the prefix to determine which API key environment variable to load: 
gemini/gemini-2.5-flash   →  reads GEMINI_API_KEY
gpt-4o                    →  reads OPENAI_API_KEY  (no prefix needed for OpenAI)
groq/llama3-70b-8192      →  reads GROQ_API_KEY
mistral/mistral-large-latest  →  reads MISTRAL_API_KEY
When the model string has no / prefix, NoteWise falls back to prefix-matching on the model name itself (gpt-* → OpenAI, claude-* → Anthropic, gemini-* → Gemini, etc.).
Gateway prefixes azure, openrouter, and vercel_ai_gateway are not supported. NoteWise will not attempt to resolve an API key for models with those prefixes.

​
Provider details

Gemini is the default provider. Gemini 2.5 Flash offers a generous free tier that covers most personal use without a billing account.Get a free API key at aistudio.google.com/app/apikey.Recommended model strings:
gemini/gemini-2.5-flash      (default — fast and free tier available)
gemini/gemini-2.5-pro        (higher quality, paid)
Vertex AI is also supported using the same GEMINI_API_KEY:
vertex/gemini-2.5-flash
vertex_ai/gemini-2.5-pro
All current OpenAI chat and reasoning models work. The model string can be used with or without the openai/ prefix.Recommended model strings:
gpt-4o
gpt-4o-mini
o1
o3-mini
Reasoning models (o1, o3, o4 series) are detected automatically and temperature is not forwarded to them, as those models do not accept a temperature parameter.
Claude 3.5 models offer strong performance on long-context tasks like transcript summarization.Recommended model strings:
claude-3-5-sonnet-20241022
claude-3-5-haiku-20241022
Groq’s LPU inference delivers very fast response times, which reduces end-to-end processing time for chunked transcripts.Recommended model strings:
groq/llama3-70b-8192
groq/llama-3.1-70b-versatile
The groq/ prefix is required.
Grok models from xAI. The xai/ prefix is required.Recommended model strings:
xai/grok-2
Mistral AI models. The mistral/ prefix is required.Recommended model strings:
mistral/mistral-large-latest
mistral/mistral-small-latest
Cohere Command models. No prefix is required for command-r-plus.Recommended model strings:
command-r-plus
command-r
DeepSeek models. The deepseek/ prefix is required.Recommended model strings:
deepseek/deepseek-chat

​
Configuring your API key

The interactive wizard prompts you to choose a provider and paste your API key, then writes it to ~/.notewise/config.env: 
notewise setup

​
Option 2: Edit config.env directly

Add the key for your chosen provider to ~/.notewise/config.env:
config.env
GEMINI_API_KEY=AIza...
DEFAULT_MODEL=gemini/gemini-2.5-flash

​
Option 3: Environment variable

Export the key in your shell before running NoteWise. Environment variables override config.env: 
export GEMINI_API_KEY=AIza...
notewise process "https://youtube.com/watch?v=VIDEO_ID"

​
Selecting a model per run

Use --model to override DEFAULT_MODEL for a single invocation without changing your config: 
# Use Claude for this run only
notewise process "https://youtube.com/watch?v=VIDEO_ID" --model claude-3-5-sonnet-20241022

# Use Groq for fast processing of a large playlist
notewise process "https://youtube.com/playlist?list=PLAYLIST_ID" --model groq/llama3-70b-8192
The --model flag takes the same LiteLLM model string format as DEFAULT_MODEL.

Build docs developers (and LLMs) love

Get started for freeTalk to us