Skip to content

Why Spillway uses local AI

The basic idea

Spillway uses artificial intelligence to help sort email, summarize messages, and identify things that may need your attention.

For the AI to do that work, the message has to be processed somewhere.

There are two main options:

  • Local AI runs on your own computer.
  • Cloud AI runs on computers operated by another company and reached through the internet.

Spillway is designed to make local AI a practical default, especially for email that may contain private, confidential, or institutionally restricted information.

What does “local” mean?

When Spillway uses a local AI model, the text being analyzed is sent to an AI service running on your Mac, such as Ollama.

In the normal local setup, the message does not need to be sent to OpenAI, Anthropic, Google, or another cloud AI company.

This is similar to the difference between editing a document on your own computer and uploading it to an online service for processing.

Installing Ollama

Ollama is the software Spillway uses to run compatible AI models directly on your Mac.

1. Download Ollama

Visit ollama.com and download the macOS version.

Open the downloaded application and follow the installation prompts. macOS may ask for permission the first time Ollama runs.

2. Confirm that Ollama is running

After installation, Ollama should run in the background. You may see its icon in the macOS menu bar.

Spillway normally connects to Ollama at the local address:

http://localhost:11434

“Localhost” means your own computer. Spillway should not need to send local-AI requests anywhere else.

3. Install a model

Open the Terminal application on your Mac and run:

ollama pull llama3.2:3b

This downloads a relatively compact general-purpose model suitable for initial testing on many Macs.

The download may take several minutes and requires free storage space. Larger models may produce better results but also use more memory, storage, and processing power.

4. Test the model

In Terminal, run:

ollama run llama3.2:3b

Type a brief message and press Return. When the model responds, enter:

/bye

to exit.

5. Configure Spillway

In Spillway:

  1. Open Settings.
  2. Open the AI or AI Provider settings.
  3. Select Ollama or Local AI.
  4. Confirm the server address is http://localhost:11434.
  5. Select the model you installed, such as llama3.2:3b.
  6. Use the available connection or model test before enabling it for email processing.

The exact labels may change during pre-alpha development, but the core setup remains the same: Ollama runs on your Mac, Spillway connects to the local service, and you choose which model and accounts may use it.

Troubleshooting Ollama

Spillway cannot connect

Confirm that Ollama is open and running. Restart Ollama, then reopen Spillway and test the connection again.

You can also open Terminal and run:

ollama list

If Ollama is working, this should show the models installed on your Mac.

The model does not appear in Spillway

Make sure the model finished downloading. Run:

ollama list

and use the exact model name shown there.

Processing is slow

Local AI performance depends heavily on the Mac's processor and memory. A smaller model may work better on older or lower-memory systems. Close other memory-intensive applications while testing.

You want to remove a model

Run:

ollama rm llama3.2:3b

Replace the name with the exact model you want to remove.

Institutional or managed Macs

Your organization may restrict installing software or running local services. Check applicable institutional policies before installing Ollama or processing work email through any AI system.

Why does this matter for email?

Email often contains much more sensitive information than people realize. A mailbox may include:

  • conversations with family and friends;
  • student or employee information;
  • unpublished research;
  • legal, clinical, or financial discussions;
  • travel plans and calendar details;
  • account information, private links, or attachments; and
  • messages that are harmless individually but revealing when combined.

Keeping AI processing on the same computer can reduce how much of that information is sent to outside services.

Is cloud AI unsafe?

Not necessarily. Cloud AI services are useful, widely used, and may offer larger or more capable models.

The important point is that using cloud AI means selected information must leave your computer and be processed under that provider's terms, policies, account settings, and security practices.

Spillway does not assume that every message, account, or workplace has the same rules. It is being designed so the user can choose the appropriate processing method rather than having one choice silently applied everywhere.

Other reasons to use local AI

Privacy is not the only benefit.

Local AI can also:

  • continue working without an internet connection after the model is installed;
  • avoid per-message cloud processing charges;
  • give the user more control over which model is installed and when it changes; and
  • remain available even when a cloud provider has an outage, changes a product, or changes account access.

There are tradeoffs. Local models use your computer's memory, storage, and processor. They may be slower or less capable than the largest cloud models, particularly on older hardware.

Do I need to understand AI models?

No. Spillway's goal is to make the choice understandable without requiring technical expertise.

The setup and settings should explain, in ordinary language:

  • where email content will be processed;
  • whether it will leave the computer;
  • which accounts are allowed to use that option; and
  • what the likely privacy, performance, and cost tradeoffs are.

A brief technical note

The kind of AI used for these tasks is often called a large language model, or LLM. Ollama is software that makes it possible to run compatible models locally.

Most users should not need to select a model based on technical specifications alone. Spillway intends to provide sensible recommendations based on the computer, the task, and the user's privacy choices.

Spillway's principle

Some users will choose local AI for everything. Others may use cloud AI for ordinary personal mail while keeping work, research, student, legal, or clinical accounts local-only.

Spillway is being built so that these decisions belong to the user and can differ by account. A single inbox view should not silently erase those boundaries.