Run AI locally with Ollama (free) in Interviewdom
Install Ollama on Windows, download a Gemma 4 model and connect it to Interviewdom. No key, no API fees; answers stay on your PC.
- Updated November 19, 2026
- Medium
- About 20 min
- No card needed
- Needs a second key for listening
1. At a glance
- What you get: answers and screenshot reading from a model running on your own PC. Nothing is sent to an AI company.
- Cost: $0 in API fees.
- No key needed.
- Listening: not supported in Interviewdom yet with Ollama; add a Groq key for listening (that audio does go to Groq).
- Trade-offs: needs a strong PC; answers are usually slower than cloud providers, especially without a good graphics card (GPU).
- Time needed: about 20 minutes (mostly the model download).
2. Check your PC
Ollama's Windows requirements:
- Windows 10 22H2 or newer (Home or Pro).
- NVIDIA graphics: driver 551.61 or newer. AMD Radeon: a ROCm v7-capable driver or a Vulkan-capable driver.
- Disk space for the model: Gemma 4 E4B is about 6.6–9.5 GB; Gemma 4 31B is about 19–20 GB.
3. Install Ollama
- Download OllamaSetup.exe for Windows.
- Run it. It installs in your user account and does not need administrator rights.
- Check it works: open PowerShell and type
ollama. You should see a list of commands.
Models are saved in %HOMEPATH%\.ollama. To store them elsewhere, set the OLLAMA_MODELS user environment variable and restart Ollama.
4. Download a model (no API key)
- In PowerShell, run:
ollama pull gemma4:e4b
- Wait for the download to finish.
There is no API key. Ollama listens on your PC at http://localhost:11434.
5. Pick a model
All Gemma 4 sizes can read screenshots.
| Model | Download size | Best for |
|---|---|---|
Gemma 4 E4B (gemma4:e4b) – recommended | about 6.6–9.5 GB | Most laptops with a decent GPU |
Gemma 4 12B (gemma4:12b) | about 7.7–8 GB | Better answers |
Gemma 4 31B (gemma4:31b) | about 19–20 GB | Best answers, needs a powerful PC |
Smaller models answer faster. During a live interview, speed matters more than depth.
6. Connect in Interviewdom
- Make sure Ollama is running.
- Go to Settings → AI Providers → Ollama.
- Keep the base URL
http://localhost:11434/v1. There is no key to paste. - Choose the model you pulled (for example
gemma4:e4b). - Click Test connection. You should see something like OK · 1200 ms (local models are often slower).
7. Speech-to-text
Ollama has a transcription endpoint for audio-capable models, but it currently accepts WAV files only, while Interviewdom records WebM audio. So for now, add a Groq key for listening in Settings → AI Providers → Speech-to-text.
8. Costs in practice (estimate)
- Ollama: $0 in API fees (only your electricity).
- Listening with Groq: about $0.04 per hour.
9. Troubleshooting
| Problem | What it means | Fix |
|---|---|---|
| Connection refused | Ollama isn't running | Start Ollama, then test again. |
| 404 model not found | The model isn't downloaded or the name is different | Run ollama pull gemma4:e4b and choose the exact same name. |
| Very slow answers | Model too big for your PC, or running on CPU | Use a smaller model (gemma4:e4b) and update your graphics driver. |
| Screenshot not understood | Images must be sent as data, not links | Interviewdom does this automatically; update the app if it fails. |
| Antivirus/firewall blocks it | Local port 11434 blocked | Allow Ollama in your firewall for local connections. |
10. Security and privacy tips
- With Ollama, your questions and screenshots stay on your PC.
- If you add Groq for listening, the interview audio goes to Groq.
- Keep Ollama updated from its official site only.