Before you start
- A computer: a Mac with macOS 14 Sonoma or newer, Windows 10 (version 22H2) or Windows 11, or Linux.
- Memory: 8 GB of RAM is enough for the small models; 16 GB or more lets you use the bigger ones.
- Disk space: about 5 GB free (Ollama itself plus one model).
- TransFlow already installed in Chrome.
Step 1 — Install Ollama
Mac
- Go to ollama.com/download/mac and press Download for macOS.
- Open the downloaded file and drag Ollama into the Applications folder.
- Open Ollama from Applications (or Launchpad). If the Mac asks whether you are sure, press Open.
- A small llama icon 🦙 appears in the menu bar at the top of the screen. If a chat window opens, you can close it — Ollama keeps running.
Windows
- Go to ollama.com/download/windows and press Download for Windows.
- Open the downloaded OllamaSetup.exe and press Install. It does not need administrator rights.
- Ollama starts by itself. A llama icon 🦙 appears next to the clock at the bottom-right (click ^ if you do not see it). If a chat window opens, you can close it.
Linux
- Open a terminal (Ctrl + Alt + T on Ubuntu and most other systems).
- Paste this official install command and press Enter. Type your password if asked:
curl -fsSL https://ollama.com/install.sh | shOllama now runs in the background and starts with the computer.
Step 2 — Let TransFlow use Ollama
Out of the box, Ollama only answers apps it trusts, and browser extensions are not on that list. You add them once with one command. Nothing else on your computer changes, and the setting survives restarts.
Mac
- Open the Terminal app: press ⌘ Cmd + Space, type Terminal and press Enter.
- Press Copy on the box below, click inside the Terminal window, press ⌘ Cmd + V, then Enter. No message appears — that is normal.
mkdir -p ~/Library/LaunchAgents && cat > ~/Library/LaunchAgents/setenv.OLLAMA_ORIGINS.plist <<'EOF'
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
<key>Label</key><string>setenv.OLLAMA_ORIGINS</string>
<key>ProgramArguments</key>
<array><string>/bin/launchctl</string><string>setenv</string><string>OLLAMA_ORIGINS</string><string>chrome-extension://*</string></array>
<key>RunAtLoad</key><true/>
</dict>
</plist>
EOF
launchctl unload ~/Library/LaunchAgents/setenv.OLLAMA_ORIGINS.plist 2>/dev/null; launchctl load ~/Library/LaunchAgents/setenv.OLLAMA_ORIGINS.plist- Restart Ollama: click the llama icon 🦙 in the menu bar → Quit Ollama. Then open Ollama again from Applications.
- Back in TransFlow, press Check connection. It should now show ✓.
Windows
- Open PowerShell: press the ⊞ Windows key, type PowerShell and press Enter. No need to run it as administrator.
- Press Copy on the box below, click inside the PowerShell window, press Ctrl + V (or right-click), then Enter. No message appears — that is normal.
[System.Environment]::SetEnvironmentVariable('OLLAMA_ORIGINS','chrome-extension://*','User')- Restart Ollama: right-click the llama icon 🦙 next to the clock → Quit Ollama. Then open Ollama again from the Start menu.
- Back in TransFlow, press Check connection. It should now show ✓.
Linux
- In a terminal, paste this command with Ctrl + Shift + V and press Enter. Type your password if asked — the letters stay invisible while you type.
sudo mkdir -p /etc/systemd/system/ollama.service.d && printf '[Service]\nEnvironment="OLLAMA_ORIGINS=chrome-extension://*"\n' | sudo tee /etc/systemd/system/ollama.service.d/transflow.conf >/dev/null && sudo systemctl daemon-reload && sudo systemctl restart ollama- The command restarts Ollama for you. Back in TransFlow, press Check connection. It should now show ✓.
The Linux command is for the standard install from step 1, which runs Ollama as a background service.
Step 3 — Download a translation model
- In Settings → Local Ollama, step 3, click a model. Not sure? Take the one marked Recommended.
- Press Download and wait for the bar to fill. A 3 GB model takes roughly 5–15 minutes on a typical home connection. You can keep using the computer meanwhile.
| Model | Download | Good for |
|---|---|---|
| TranslateGemma 4B — Recommended | ~3.3 GB | Best quality for its size, 55 languages. Fine on 8 GB of RAM. |
| Hy-MT2 1.8B | ~1.5 GB | Smallest and fastest translator, about 30 languages. For older or low-memory computers. |
| Qwen 3.5 (2B / 4B / 9B) | 1.7 – 6.6 GB | General AI models: 201 languages, can also apply Tone, explain and read text in pictures. |
| TranslateGemma 12B | ~8.1 GB | Most accurate. Needs 16 GB of RAM or more. |
Rule of thumb: with 8 GB of RAM choose a model under 3.5 GB; with 16 GB or more any of them works. Part B on the same page is only for translating text inside pictures — skip it unless you need that.
Step 4 — Switch it on and try it
- In step 4, turn on Use Ollama for translation.
- Press Try a sample translation. The very first run can take up to about 30 seconds while the model loads into memory; after that it is much faster.
- Optional: to use Ollama first for everything, open the Translation tab and choose Private / offline.
If something does not work
“Cannot find Ollama”
Ollama is not open. Open it from Applications (Mac) or the Start menu (Windows) and look for the llama icon 🦙, then press Check connection again. If you changed the Ollama address box, set it back to http://127.0.0.1:11434.
“Ollama is blocking TransFlow” even after the unlock
- Ollama must be fully quit and reopened — closing its window is not enough. Use Quit Ollama from the llama icon, then open it again.
- On Windows, make sure you pasted into PowerShell, not Command Prompt.
- Paste the command using the Copy button instead of selecting the text by hand, so nothing is missing.
- Still stuck? Restart the computer once — that always reloads the setting.
The download fails or stops
Check your internet connection and free disk space, then press Download again — it continues where it stopped.
Translation is slow, or the computer gets slow
The first translation after a break is slow because the model is loading. If it stays slow, pick a smaller model. A loaded model uses memory; Ollama releases it after 30 minutes without use, or press Unload under RAM in use in the same settings page.
Tone has no effect
Translation-only models (TranslateGemma, Hy-MT2) ignore Tone. Choose a Qwen model if you want Tone and the Explain feature on your own computer.
Undo the unlock or remove Ollama
The unlock lets browser extensions installed on this computer talk to Ollama. Websites and other computers still cannot reach it. To take the unlock back, run the command for your computer and restart Ollama:
Mac (Terminal)
launchctl unload ~/Library/LaunchAgents/setenv.OLLAMA_ORIGINS.plist; rm -f ~/Library/LaunchAgents/setenv.OLLAMA_ORIGINS.plist; launchctl unsetenv OLLAMA_ORIGINSWindows (PowerShell)
[System.Environment]::SetEnvironmentVariable('OLLAMA_ORIGINS',$null,'User')Linux
sudo rm -f /etc/systemd/system/ollama.service.d/transflow.conf && sudo systemctl daemon-reload && sudo systemctl restart ollamaTo remove Ollama completely: on Mac, quit it and move Ollama from Applications to the Trash; on Windows, uninstall it from Settings → Apps. Downloaded models live in the .ollama folder in your home folder — delete it to free the space.