FanumTag - Local-First AI File Renamer

FanumTag is a local-first desktop rename workspace built with Tauri (Rust) + SolidJS (TypeScript). It scans a folder into a visual queue, uses local AI to suggest descriptive filenames, and applies safe bulk renames without uploading your files.

---

Installation

FanumTag is currently distributed for 64-bit Windows.

MSI release

Download fanumtag_1.2.0_x64_en-US.msi and its .sha256 file from the GitHub release. Verify it in PowerShell before installing:

(Get-FileHash .\fanumtag_1.2.0_x64_en-US.msi -Algorithm SHA256).Hash.ToLower()
Get-Content .\fanumtag_1.2.0_x64_en-US.msi.sha256

The values must match. This MSI contains FanumTag and its native runtime, but not the large GGUF weights. After installation, open Settings → Runtime Settings, select an existing Qwen .gguf model and compatible MMProj .gguf, click Save Configuration, and then start the runtime. For the default sharded Qwen model, select its first shard and keep its second shard in the same directory.

Complete NSIS package

A complete NSIS package with the default weights has this layout:

FanumTag-windows-x64/
├── fanumtag_1.2.0_x64-setup.exe
└── weights/
    ├── Qwen3-VL-4B-Instruct-Q4_K_M.gguf-00001-of-00002.gguf
    ├── Qwen3-VL-4B-Instruct-Q4_K_M.gguf-00002-of-00002.gguf
    └── mmproj-F16.gguf

1. Download the complete Windows package and extract it if it is archived.
2. Keep fanumtag_1.2.0_x64-setup.exe beside the weights directory.
3. Run the installer. It verifies all three model files and copies them into the installed application automatically.
4. Open Settings → Runtime Settings and leave Qwen Model Path and MMProj Model Path empty to use the bundled defaults.
5. Start the runtime. The first Qwen shard automatically discovers the second shard in the same directory.

Do not run a detached copy of the installer: installation stops with a clear error if its weights directory is missing. The complete package requires approximately 3.44 GiB of disk space before installation. A release without a Windows package contains source code only and must be built using the development instructions below.

Using custom weights

In Settings → Runtime Settings, use Browse to select another compatible .gguf Qwen model or multimodal projector, then click Save Configuration. For a sharded model, select the first shard and keep every shard together. Clear a path and save to return to the bundled default. FanumTag checks that both selected files exist when the runtime starts and reports the missing path if either cannot be found.

---

Runtime Architecture

• Rust owns a singleton local runtime manager.
• The manager starts one bundled llama-server.exe process from src-tauri/lib.
• Vision weights are loaded from src-tauri/weights:
  • Qwen3-VL-4B-Instruct-Q4_K_M.gguf-00001-of-00002.gguf
  • Qwen3-VL-4B-Instruct-Q4_K_M.gguf-00002-of-00002.gguf
  • mmproj-F16.gguf
• Complete Windows NSIS packages include the installer and a weights sidecar directory. The installer verifies and copies the model shards and projector into the installed app by default; keep the release folder together when distributing it. Users can override either installed path from Runtime Settings when they want another compatible model or projector. For a sharded custom model, select its first shard; llama.cpp discovers the remaining shards automatically.
• Frontend communicates through Tauri commands/events.

---

Core Commands

• runtime_get_status
• runtime_get_config
• runtime_update_config
• runtime_start
• runtime_stop
• runtime_cancel_batch
• runtime_generate_batch
• apply_renames

---

Features

• Workspace queue with folder quick access, subfolder browsing, thumbnails, search, sorting, filtering, and pagination
• Batch suggestions for images, videos, and text files, with deterministic fallback handling
• Live generation progress with per-file status, ready/pending counts, and a Stop control
• Selection tools for selecting all, selecting a page, selecting ready items, inverting a page, and keeping ready items
• Safe native renames with collision handling and Windows-name validation
• Runtime settings for host, port, threads, GPU layers, context size, request timeout, and auto-start
• Runtime health checks for bundled inference, Whisper, and FFmpeg dependencies

---

Screenshots

Empty workspace

Start by selecting a folder. The workspace provides quick access to common folders and keeps the selection and pagination panels ready for the queue.

Loaded queue

Review thumbnails and pending files before generating suggestions. The queue supports search, sorting, file-type filtering, subfolders, and page navigation.

Generating suggestions

Generation progress is shown in the workspace footer and in the per-file status column. You can stop an active batch at any time.

Runtime settings

Configure the local runtime and check the health of its bundled dependencies from the Settings view.

---

Development

pnpm install
pnpm tauri dev

Development expects the llama.cpp runtime files under src-tauri/lib and the three GGUF files shown above under src-tauri/weights. These large local assets are ignored by Git and are not included in GitHub's automatic source archives.

Checks

pnpm build
pnpm serve

Notes

• Keep runtime and model paths local for privacy.
• No cloud dependency is required for default workflow.