Renaming files with a local LLM, your own API key, or neither
I build Zush, a desktop app that renames files by reading what's inside them. It can call three kinds of model: our hosted one, a provider you pay yourself, or a model running on your machine through Ollama or LM Studio.
I build Zush, a desktop app that renames files by reading what's inside them. It can call three kinds of model: our hosted one, a provider you pay yourself, or a model running on your machine through Ollama or LM Studio. This post is the practical side: what each route sends where, how to wire up the local ones, and how to pick per folder.
Competitor details below come from vendor pricing and docs pages checked on November 20, 2026.
What actually gets sent
Renaming a file with a vision or language model looks like this: the app extracts something the model can read, sends it with a prompt, and gets back a short title that the naming template turns into a filename.
What gets extracted depends on the file. For a PDF it's the text, or recognized text if it's a scan. For an image it's a preview. For audio and video it's a transcript of the speech plus frames and metadata. The original file stays where it is in every mode.
The routes differ in who receives that payload:
| Route in Zush | Where the model runs | Who bills you | Needs internet |
|---|---|---|---|
| Zush Cloud AI | Zush's server | Included in Zush | Yes |
| BYOK | Google, OpenAI, Anthropic or OpenRouter | Your provider | Yes |
| Ollama | Your computer | Nobody | Only for the first model download |
| LM Studio | Your computer | Nobody | Only for the first model download |
With Ollama or LM Studio, speech recognition for audio and video also runs locally.
Wiring up Ollama
Zush talks to Ollama on its default address, http://127.0.0.1:11434.
- In Zush, open Settings, then AI Modes, then Ollama.
- If Ollama isn't installed, use the Download Ollama button, install it, and come back.
- Press Start Ollama if the server isn't running.
- If you have no models, press Download next to the suggested model. Zush shows progress and selects it when done.
- Press Test, then rename three or four files you know well before pointing it at a big folder.
If you already run Ollama with your own models, Zush shows a model picker instead. Pick a vision-capable model. A text-only model can't look at image previews, so screenshots and photos will fail or come back vague.
Wiring up LM Studio
LM Studio exposes an OpenAI-compatible server, by default on http://127.0.0.1:1234.
- Install LM Studio and download a vision-capable model.
- In Zush, choose LM Studio in AI Modes. Since Zush 3.23.2, the Start LM Studio button also starts the LM Studio server for you.
- Refresh the model list. Zush only lists models that LM Studio reports as vision-capable.
- Press Test and try a small batch.
When local is the wrong choice
Local models are slower on modest hardware, and they need memory to load. If the machine struggles, try a smaller model. If that hurts quality too much, local probably isn't the right route for that machine, and that's fine. The cloud routes run the model on a server, so a basic laptop handles them without trouble.
For sensitive folders, local is the safest default. For a pile of meme screenshots, it's overkill.
BYOK in two minutes
Create a key in Google AI Studio, the OpenAI platform, the Anthropic console or OpenRouter. In Zush, choose BYOK in AI Modes, pick the provider, paste the key, save, and press Test. Zush loads the models your key can use. Your provider bills you under its own pricing, so check rate limits before you run a thousand files.
A sane per-folder setup
What I'd do on a work machine:
- Monitor a downloads or scans folder with Zush Cloud AI, because it needs no setup and works on any laptop.
- Switch to Ollama or LM Studio before renaming anything with client or medical data.
- Use BYOK if your company already pays for an API and wants usage on that bill.
Templates, the preview window and the undo history stay the same whichever route you pick.
Other apps that do this
If you'd rather not run Ollama at all but still want local, RenameClick ships its own model. It's about a 4 GB download, and the vendor recommends 16 GB of RAM. FilesDesk can also talk to vLLM, if that's your stack. AI Renamer, FilesDesk and NameQuick sell one-time licenses that run on your key or a local model only; their own cloud is a subscription or credits.
Zush's one-time license is $48 and includes the hosted cloud, BYOK and both local routes on 3 devices, Mac and Windows mixed. There's also a $10 monthly plan. You can try every route on 50 free renames without creating an account.
Docs: Ollama Β¬Π Docs: LM Studio Β¬Π Docs: BYOK
If you've run file renaming against a local model, I'm curious which model gave you the best names for scanned documents. Comments are open.
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes β full credit and traffic to the original publisher.