# Connect Ollama, LM Studio, or cloud AI models | Frog

Canonical page: https://frog.beffa.xyz/docs/providers/
Updated: 2026-10-10

Models → Providers connects Frog to local servers, cloud APIs, or compatible endpoints. Models → Installed tools uses supported coding CLIs with their existing login. Downloaded models that run inside Frog live under Models → Local models.

These guides describe Frog 1.0.42, the published download as of October 29, 2026. Download links lead to the latest published release.

## Add a provider connection

Use model IDs your account or server supports. Changing a display name does not change the API ID. API keys are stored in macOS Keychain rather than portable configuration files. Testing a text connection sends a short test message.

Provider connections are separate from Mr. Usage logins. The usage dashboard reads credentials owned by your coding tools; adding a writing provider here does not configure a usage account.

1. Open Models → Providers and add a connection.
2. Name the connection, choose the service, and enter its endpoint and credentials where required.
3. Add the models available to that service. Each model has an API ID, a display name, and a text or speech-to-text category.
4. Test and save the connection, then select one of its models in Writing or Dictation.

## Use Ollama on your Mac

Frog connects to the existing Ollama server; it does not install or manage the server. A localhost endpoint processes requests on your Mac. A remote Ollama endpoint processes them on the machine hosting that endpoint.

1. Install and start Ollama, then download a compatible text model through Ollama.
2. Add an Ollama connection in Frog. The usual local endpoint is http://localhost:11434.
3. Choose Find installed models, select the models you want, test the connection, and save.

## Use LM Studio on your Mac

Model discovery asks the configured server for its model list. It does not send your writing or microphone recordings.

1. Install LM Studio, download a model, and start its local server in the Developer section.
2. Add an LM Studio connection in Frog. The usual local endpoint is http://localhost:1234/v1.
3. Find installed models, test the connection, and save. Supply an API token if your server requires one.

## Connect a cloud service

Frog supports OpenAI, Anthropic Claude, Google Gemini, and compatible API endpoints. Cloud services use their own credentials, model availability, pricing, and data policies. When you run a cloud-backed writing rule, the selected text and instructions are sent to that service.

External speech-to-text is available through supported OpenAI transcription, Gemini audio input, and compatible transcription endpoints. Selecting an external speech model sends the recording for transcription. Not every text provider offers speech-to-text.

## Use an installed coding tool’s text models

Frog invokes the supported CLI in the background for text transformations, with restricted tool access and cancellation. It does not automate a desktop chat window, install the tool, or sign you in. Codex support is limited to verified 0.159.x interfaces; other minor versions remain unsupported. A detected login is not a guarantee that every model is available to the account.

The portable tool-default model ID is @default; Frog omits the CLI model argument for that choice. Executable paths, login state, and tokens stay out of configuration. The CLI owns authentication and any credential refresh; Frog does not copy its credentials into Frog’s Keychain. Requests still use the selected tool’s provider and account policies, so an installed CLI does not imply offline inference.

OpenCode retains its own CLI session history even when Frog’s result history is off. Frog’s temporary request files are removed after use. Isolated tool invocation does not import custom provider aliases or project customization; use an available model or the tool default.

1. Install and sign in to Claude Code, Codex, or OpenCode through that tool’s own setup.
2. Open Models → Installed tools. Frog shows installation, login, and supported-interface status separately.
3. Choose an available text model or the tool default and select Use model. Frog creates or updates the connection without an endpoint or Frog API-key form.
4. Select that model in a writing or translation rule, or as the text model for dictation cleanup.

## Choose a model for each rule

Each rule stores its own model selection; there are no global model defaults that reroute configured rules. You can mix local and cloud-backed rules. Deleting a provider connection clears affected rule selections rather than silently switching them to another provider.

## Related guides

- [Local models](https://frog.beffa.xyz/docs/local-models/)
- [Writing rules](https://frog.beffa.xyz/docs/writing-rules/)
- [Privacy and offline use](https://frog.beffa.xyz/docs/privacy/)

## Source documentation

- [Provider endpoints and supported model categories](https://github.com/t1llo/frog/blob/main/docs/providers.md)
- [Installed-tool integration and support boundaries](https://github.com/t1llo/frog/blob/main/docs/local-tool-providers.md)