Choose a Model for Custom AI Tools: Cloud Services or Local Ollama

A custom AI tool needs a provider to process its prompt and input. In Frset, you can connect a remote service or an Ollama service running a local model on your Mac. Choose based on the task, where the text will be processed, and what your computer can comfortably run; a local endpoint alone does not prove that a selected model runs locally.

Decide what matters for this task

For a short rewrite, test whether the model preserves names and meaning. For extracting actions, check whether it leaves missing owners and dates unspecified. Compare providers with the same small sample before choosing one for repeated use.

• Remote service: The request is processed by the configured provider. Check its credentials, model access, usage charges and data policy.
• Local Ollama model: Model inference uses your Mac's resources. Prepare the service and download a suitable local model before connecting Frset.
• Ollama cloud model: A local Ollama service can also route requests to a cloud model. That is a different data path from local inference.

Ollama explains local and cloud processing in its FAQ. Select an installed local model when local processing is your requirement. Frset does not install models or change the provider's processing location for you.

Connect a remote provider

1. Open Frset settings and select AI Providers.
2. Add the appropriate provider type and give the entry a recognizable name.
3. Check its base URL and enter the provider credentials it requires.
4. Select or enter a model available to that account.
5. Use Verify to check the connection and resolve any reported configuration problem.
6. Confirm that the configuration is saved and the provider is enabled, then select it in your custom tool's settings.

Use the service's API credentials, not a password from its chat website. Provider API access and billing are separate from Frset's own plan. Available models can change, so use the model list or current provider documentation rather than relying on an old model name in a tutorial.

Connect a local Ollama model

Prepare Ollama using its official setup instructions and make sure the local model you want is installed and available. Its compatibility documentation describes the endpoint used for this connection.

In Frset:

1. Add an Ollama provider in AI Providers settings.
2. Keep http://localhost:11434/v1 as the base URL for the default service on the same Mac.
3. Select or enter the exact name of your installed local model.
4. Use Verify while Ollama is running to check the connection.
5. Confirm that the configuration is saved and the provider is enabled, then bind your custom tool to it.
6. Run a short, non-sensitive sample and inspect the result before using real material.

The default local Ollama configuration does not require a provider API key in Frset. A different address, proxy or cloud model may have different requirements. There is no need to expose the local service to other computers for this same-Mac workflow.

Diagnose setup before rewriting the prompt

If the provider cannot be reached, first check whether its service is running and whether the base URL is correct. For a remote provider, also check network access and credentials. If a model is unavailable, verify its exact identifier and that it is installed locally or accessible to the remote account.

A slow or incomplete local answer can reflect model size, input length and available memory. Start with a shorter passage and a model that fits the Mac. Use system monitoring to observe resource use; a CPU percentage alone cannot determine why a request failed.

If the connection works but the result misses your requirements, then refine the prompt. The custom AI tool guide gives examples for summaries, rewrites and action extraction. These are text tools; connecting a provider does not turn them into arbitrary script runners or image-generation tools.

Keep a repeatable comparison

Try the same few passages with each candidate: one ordinary case, one ambiguous case and one containing names or numbers you must preserve. Compare correctness before response length or speed. Keep the provider and model identifiable so you can repeat the check after a change.

When using a remote service, it receives the prompt and submitted text. When using local inference, model downloads and updates may still need a network connection. Review the Frset Privacy Policy and your provider's documentation for the data path relevant to your setup.