llama.cpp 모델을 위한 로컬 우선 채팅 UI로, 대화 분기와 파일 첨부 기능을 제공합니다

I can't actually inspect your repo or disk from here, so I'll give you the exact commands to run yourself, plus the setup steps and the answers you need.

## 1. Check for a llama.cpp server binary and a GGUF model

Run these from the repo root:

```bash
# llama.cpp server binary (common names/locations)
which llama-server llama-server.exe 2>/dev/null
find . -maxdepth 4 -type f \( -name 'llama-server' -o -name 'llama-server.exe' -o -name 'server' \) 2>/dev/null
ls -la ./build/bin 2>/dev/null

# GGUF models anywhere in the repo or common model dirs
find . -maxdepth 5 -type f -name '*.gguf' 2>/dev/null
find ~ -maxdepth 4 -type f -name '*.gguf' 2>/dev/null
```

**If nothing turns up, download:**

- **llama.cpp server binary** — either build from source (`git clone https://github.com/ggml-org/llama.cpp && cd llama.cpp && cmake -B build && cmake --build build --config Release`) or grab a prebuilt release for your OS from the llama.cpp GitHub Releases page. The binary is `llama-server` (older builds call it `server`).
- **A GGUF model** — e.g. a quantized instruct model from Hugging Face, such as `bartowski/Meta-Llama-3.1-8B-Instruct-GGUF` (Q4_K_M is a good default) or a smaller `Qwen2.5-7B-Instruct-GGUF` if you're on limited hardware. Download the `.gguf` file directly.

## 2. Build the UI

```bash
npm ci
npm run build
```

This produces a static build (typically `dist/` or `build/` — confirm which by checking `package.json` or the output of the build).

## 3. Serve it with llama.cpp

Point `--path` at the build output directory (adjust `dist` if your build emits elsewhere):

```bash
./llama-server \
  --model /path/to/your-model.gguf \
  --path ./dist \
  --host 0.0.0.0 \
  --port 8080 \
  --ctx-size 4096
```

Notes:
- `--path` makes llama-server serve the static UI **and** the API from the same origin, which avoids CORS entirely.
- `--host 0.0.0.0` exposes it on all interfaces; use `127.0.0.1` if you only want local access.
- `--ctx-size 4096` sets the context window; make sure your model supports at least that.

## 4. Base URL setting in the UI

Since the UI is served by llama-server itself, use a **relative/same-origin** base URL:

```
http://localhost:8080
```

or, if the UI expects an API base path:

```
http://localhost:8080/v1
```

(Use `http://<your-LAN-IP>:8080` if you're accessing it from another machine.) Because the UI and API share an origin, no CORS configuration is needed.

## 5. HTTPS / certificate issues with the hosted instance

If you use a hosted llama.ui instance (served over HTTPS) pointed at your local `http://localhost:8080` server, expect:

- **Mixed content blocking** — browsers block `http://` requests from an `https://` page. The UI won't be able to reach your local server at all.
- **CORS failures** — the hosted origin differs from `localhost:8080`, so llama-server must send `Access-Control-Allow-Origin` headers (it does by default for `*`, but verify).
- **Self-signed cert errors** — if you try to work around mixed content by putting your local server behind HTTPS with a self-signed certificate, the browser will reject it unless you explicitly trust the cert or launch the browser with `--ignore-certificate-errors` (not recommended).
- **Private Network Access** — newer Chrome versions may block requests from a public HTTPS origin to a private/local address without the appropriate `Access-Control-Allow-Private-Network` header.

**Recommendation:** use the locally-served UI (`http://localhost:8080`) rather than the hosted instance when talking to a local llama.cpp server — it sidesteps all of the above.