> ## Documentation Index
> Fetch the complete documentation index at: https://docs.snaply.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# AI Models

> Choose between Snaply's built-in local model, hosted cloud models, and OpenAI-compatible endpoints.

**Think of this as Snaply's brain switch.**

[Writing Assistant](/writing-assistant/overview) uses the model you select in Settings, and that choice affects more than just text rewrites.

The same model setting is also used for:

* Meeting titles
* Meeting note generation
* Meeting Notes Ask AI conversations

## Local Model

Snaply includes one local Writing Assistant model:

* **Daily Assistant (Gemma 4)** (`Gemma 4 E4B`, about `5.25 GB`)

Local use means:

* **runs on your Mac**
* **works offline after setup**
* **keeps text local to your machine**
* Requires macOS 14 or newer
* Requires more than 8 GB of system memory for the local Writing Assistant flow

## Cloud Models

If you prefer a hosted model, Snaply now includes a much larger cloud catalog. Recommended models are there for quick setup; search is there when you know exactly what you want.

The same model picker also includes **OpenAI-compatible endpoint** for Ollama, LM Studio, LiteLLM, local servers, proxies, and other OpenAI-style APIs.

| Provider                   | Examples in the recommended list                                                                   |
| -------------------------- | -------------------------------------------------------------------------------------------------- |
| OpenAI                     | GPT-5.1, GPT-5.1 Chat, GPT-5 Mini, o4-mini                                                         |
| Anthropic                  | Claude Sonnet 4.6, Claude Sonnet 4.5 (latest), Claude Haiku 4.5 (latest), Claude Opus 4.5 (latest) |
| Google Gemini              | Gemini 3.1 Pro Preview, Gemini 3 Flash Preview, Gemini 2.5 Pro, Gemini 2.5 Flash                   |
| OpenRouter                 | Claude Sonnet 4.5, GPT-5.1, Gemini 3 Pro Preview, Grok 4.1 Fast                                    |
| xAI                        | Grok 4.1 Fast, Grok 4, Grok 3 Fast                                                                 |
| Groq                       | GPT OSS 120B, Llama 3.3 70B Versatile, Compound                                                    |
| Mistral                    | Mistral Large (latest), Magistral Medium (latest), Codestral (latest)                              |
| OpenAI-compatible endpoint | A custom base URL and model name that you configure                                                |

Hosted cloud use means:

* Uses an internet connection
* Uses your own API key for the selected provider
* Sends the request text to that provider
* Gives you more model choice and extra power for larger reasoning tasks

OpenAI-compatible endpoints are different: Snaply sends the request to the endpoint URL you configure. That can be local, like Ollama or LM Studio on `localhost`, or remote, like a LiteLLM gateway, company gateway, or hosted proxy.

<Note>
  If Writing Assistant uses the built-in local model, selected text, meeting titles, generated meeting notes, and Ask AI requests stay local. Built-in cloud providers send that request text to the selected provider. OpenAI-compatible endpoints send it to the URL you configured, which can be local or remote.
</Note>

## OpenAI-Compatible Endpoints

Use **OpenAI-compatible endpoint** when you want Snaply to talk to a local server, LiteLLM proxy, company gateway, or hosted provider that exposes an OpenAI-style API.

You configure:

* **Base URL**, such as `http://localhost:11434/v1`
* **Model name**, such as `llama3.1:8b`
* **API key**, when the endpoint requires one

This is the power-user path: you bring the endpoint, Snaply brings the Writing Assistant and Meeting Notes workflow.

It works well for local model tools too:

| Tool                                                          | Base URL to try              | Model name                                                          |
| ------------------------------------------------------------- | ---------------------------- | ------------------------------------------------------------------- |
| [Ollama](https://docs.ollama.com/api/openai-compatibility)    | `http://localhost:11434/v1`  | The model you pulled in Ollama, such as `llama3.2` or `llama3.1:8b` |
| [LM Studio](https://lmstudio.ai/docs/developer/openai-compat) | `http://localhost:1234/v1`   | The model identifier shown by LM Studio's local server              |
| LiteLLM or any OpenAI-compatible API                          | Your server's `/v1` endpoint | The model name that endpoint expects                                |

For local tools, leave **API key** empty unless your endpoint requires a bearer token. Snaply also recognizes the common Ollama and LM Studio local ports and can suggest the correct `/v1` URL if you forget it.

## API Keys

Open **Settings -> Writing Assistant -> AI Models -> Add API key** to manage provider keys.

Snaply shows which providers already have keys saved, and keys are stored in Snaply's encrypted app vault on your Mac. If you used an older Snaply build, the app can also load previously saved keys into the newer storage.

## Choosing Between Local and Cloud

Choose **local** when you want **on-device privacy** and **offline use** after setup.

Choose **cloud** if:

* You want access to a state-of-the-art hosted model
* You prefer a specific provider or model family
* You want extra flexibility for heavier writing, reasoning, or meeting follow-up work

## Related Pages

* [Back to Writing Assistant overview](/writing-assistant/overview)
* [Set up cloud models](/writing-assistant/cloud-models)
* [Create reusable assistant modes](/writing-assistant/modes)
* [See how this affects Meeting Notes Ask AI](/meeting-notes/ask-ai)