Custom LLMs location

[LCN3-primavoce]

Description

Hi. Thank you for your work on this really cool app!

This is a bit of an edge case.

Now that there is session memory, the feature I am most excited about is built-in web-search (without setting up Docker, etc.).

I haven't been able to try it because I have not been able to use Thuki's built-in ollama server.

— I keep my OS and work on separate partitions on my MacBook's internal hard-drive. I have my various LLM models stored on the work partition (to save space on my OS partition), and Thuki doesn't see those models when I activate Thuki's built-in ollama server. Is there a way to point Thuki at the directory in which I store LLM models?

Thank you!

Steps to Reproduce

Open Thuki (any way).
Type a "/search" question — e.g. "/search what has Marco Rubio done in the past week?"

Error:
"Web search needs the built-in engine
Switch to Built-in in Settings to use /search."

Go to Thuki settings — switch to "Built-in"
--> Thuki does not see any of the local models already downloaded on my computer.

Expected Behavior

Ideally, I would be able to tell Thuki where to look for local models already downloaded to my computer.

Actual Behavior

--> Thuki does not see any of the local models already downloaded on my computer.

macOS Version

macOS Tahoe 25.6.1

Thuki Version

0.16.1

Ollama Version

0.32.1

AI Backend

Local Ollama

[quiet-node]

Hey @LCN3-primavoce thanks for the report, glad the app's been useful!

Quick fix on terms: Thuki has no built-in Ollama server. "Built-in" is a bundled llama.cpp llama-server, a separate engine from Ollama, with its own model store. So:

• Built-in only runs models downloaded via Settings → Models. It's manifest-driven, never scans folders, so .gguf files you placed on disk yourself won't appear (that's the empty library you saw, working as designed).
• Ollama is reached only over HTTP at 127.0.0.1:11434. Thuki's built-in engine never reads Ollama's files, only when you use Ollama provider that's when you can use Ollama LLM models.

So three quick questions for you:

1. What does ollama list print? If your models are there, Thuki's Ollama provider already reaches them. If not, you've got loose .gguf files: different problem.
2. What would a directory setting solve? Moving Thuki's own downloads off the OS partition, or reusing files you already have?
3. Would /search (or auto-search) on a model you already have (with Ollama provider) be the real fix for you? It only works with Built-in engine today, so knowing if that's your actual blocker helps me prioritize.

Note: /search is Built-in-only by design, unrelated to model location. Downloading a Built-in model unblocks it now, but it lands under your OS partition (~/Library/Application Support/com.quietnode.thuki/models), eating the space you're trying to save.

Again thanks for the nice words! I'm looking forward to more info from you!