---
title: "ollama: what it listens on, why it downloads models, and whether to block it · Kilobat"
url: "https://kilobat.app/processes/ollama"
description: "ollama runs large language models on your Mac and listens on port 11434. What it connects to, how to keep it local, and what breaks if you block it."
---

# What is ollama?

Ollama runs models on your own Mac. Its server listens on one local port, and it goes online mainly to pull models.

**Fix this with your agent**Copies a prompt for Claude Code, Codex or any agent that can use Terminal.

I want to know what ollama is doing on my Mac and whether its network use is normal. Read https://kilobat.app/processes/ollama.md and work through it with me. Install the Kilobat CLI if it’s missing: brew install seifip/kilobat/kilobat. Run the guide’s Terminal and kilobat checks yourself and tell me what you find. List the steps only I can do. Ask before you change any setting.

Process: `ollama` · Also seen as: `Ollama` · Made by Ollama · Applies to: Ollama on macOS 14 Sonoma and later. Checked on 7 November 2026 against Ollama's FAQ. · Checked: 7 November 2026

## Short answer

Ollama runs language models locally. It listens on 127.0.0.1 port 11434 by default and downloads models from the internet when you pull them.

Block with care

Blocking Ollama's outbound traffic stops model pulls and cloud features. Local models that you already have keep working.

Path

`/Applications/Ollama.app/Contents/MacOS/Ollama` `/opt/homebrew/bin/ollama` `/usr/local/bin/ollama`

## What it is

Ollama's FAQ says that it binds to `127.0.0.1` on port `11434` by default. You change the address with the `OLLAMA_HOST` environment variable.

It accepts cross-origin requests from `127.0.0.1` and `0.0.0.0` by default. You can add more origins with `OLLAMA_ORIGINS`.

Ollama says that it runs locally and that it does not see your prompts or data when you run models locally.

For cloud-hosted models, Ollama says that it processes prompts and responses to provide the service. It says that it does not store or log that content.

The FAQ says that you can run local-only. Set `OLLAMA_NO_CLOUD=1`, or turn on the `disable_ollama_cloud` key in `~/.ollama/server.json`, then restart. This also disables cloud models and web search.

## Why it uses the network

- **Model downloads.** `ollama pull` downloads model files. The FAQ says that Ollama uses HTTPS for these pulls, so use `HTTPS_PROXY` for a proxy.
- **The local server.** Apps and tools on your Mac send requests to port 11434. This traffic stays on the Mac unless you change the bind address.
- **Cloud models and web search.** If you use them, your prompts go to Ollama's service. Local-only mode turns these off.

## How much is normal

Outside a download, a local-only Ollama sends nothing to the internet that the FAQ describes.

Connections to `127.0.0.1` on port 11434 are usual. They come from your own apps.

Models are stored in `~/.ollama/models` on a Mac. You can change this with `OLLAMA_MODELS`.

## When it uses a lot, and why

- **Pulling a model.** Models are large files. Check the size shown in the model's listing first.
- **Exposing the server.** If you set `OLLAMA_HOST` to a non-local address, other devices can send requests and use your Mac's processor and memory.
- **Cloud models.** Prompts and replies go over the network to Ollama's service, so traffic follows how much you use them.

## Is it safe to block?

Block with care

Blocking Ollama's outbound traffic stops model pulls and cloud features. Local models that you already have keep working.

What breaks if you block it

- `ollama pull` and model updates fail.
- Cloud models and web search stop.
- Blocking port 11434 on the loopback address breaks every app that talks to your local Ollama.

Blocking means a firewall rule that refuses its connections. Don’t quit or disable a macOS process to save data: macOS starts it again, and other features depend on it.

## Reduce its use

1. To keep Ollama local, leave `OLLAMA_HOST` unset. It then binds to `127.0.0.1`.
2. To turn off cloud features, set `OLLAMA_NO_CLOUD=1` and restart Ollama.
3. Pull models on a network with room for them. Check a model's size first.
4. Behind a proxy, set `HTTPS_PROXY`. The FAQ says to avoid `HTTP_PROXY`.

## See how much it uses

1. In Terminal, list what is listening on the Ollama port.

   ```
   lsof -nP -iTCP:11434 -sTCP:LISTEN
   ```

## With Kilobat

1. In the Kilobat app list, search for `ollama`. Its row shows the live rate and the hosts.
2. Ask the same question in Terminal.

   ```
   kilobat explain ollama
   ```

The live view and the explanations are free. History in the app needs Kilobat Pro. Blocking needs Pro and the network filter: see the [network filter](https://kilobat.app/docs/blocking#filtering), [Protection](https://kilobat.app/docs/blocking#protection) and the [hotspot data saver](https://kilobat.app/save-hotspot-data).

## What isn’t documented

Ollama's FAQ does not list the hosts that it contacts for model pulls or updates, so this entry names none.

The FAQ does not say whether the desktop app checks for updates, or how often.

## Questions

**Can other computers use my Ollama?**

Not by default. Ollama binds to `127.0.0.1`. They can reach it only if you change `OLLAMA_HOST` or put a proxy in front of it.

**Does Ollama send my prompts to the cloud?**

Ollama says that it does not see prompts when you run models locally. Cloud-hosted models are different, as they process prompts in Ollama's service.

## Sources

- [Ollama Docs: FAQ](https://docs.ollama.com/faq)
- [Kilobat docs: Connections](https://kilobat.app/docs)

Checked on 7 November 2026.

## Related

- [node](https://kilobat.app/processes/node): The Node.js runtime, not part of macOS. It connects only when a script asks, such as npm installing packages, a dev server, an AI agent or an MCP server.
- [Codex](https://kilobat.app/processes/codex): OpenAI's coding agent, available as a desktop app, command-line tool and other interfaces.
