docs(release): INSTALL.md + dev-preview release notes for v0.1

v0.1 Tauri GUI ships as a developer preview that wraps the Python
langslice-harness CLI. INSTALL.md walks users through the one-time
prereq setup (Python 3.10+, `pip install -e .`, BrainGlobe atlas)
and explains the contrast with the no-prereq browser demo at
greenpolo.github.io/LangSlice.

Release-workflow notes now point at INSTALL.md and are explicit about
the dev-preview status so judges/users don't expect a fully self-
contained installer.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: greenpolo

.github/workflows/release-tauri.yml

@@ -51,17 +51,26 @@ jobs:
           tagName: ${{ github.ref_name }}
           releaseName: 'LangSlice ${{ github.ref_name }}'
           releaseBody: |
-            **LangSlice ${{ github.ref_name }}** — Tauri desktop installer.
+            **LangSlice ${{ github.ref_name }} — developer preview**
 
-            Pick whichever installer you prefer:
-            - `.exe` — friendly NSIS installer, double-click to install.
-            - `.msi` — Microsoft Installer format (for IT-managed deployment).
+            Windows desktop app for the LangSlice brain-slice AP-position
+            pipeline. **Read [INSTALL.md](https://github.com/greenpolo/LangSlice/blob/main/INSTALL.md)
+            before downloading** — this build requires Python 3.10+ and the
+            `langslice-harness` package on the host machine. For the
+            no-prereq option, use [the browser demo](https://greenpolo.github.io/LangSlice).
+
+            ## Downloads
+            - `.exe` — NSIS installer, double-click to install.
+            - `.msi` — Microsoft Installer format (IT-managed deployment).
 
             On first launch Windows SmartScreen may warn that the installer is
             unsigned — click **More info → Run anyway**.
 
-            Local LangSlice-Gemma 4 model card:
-            https://huggingface.co/greenpolo/langslice-gemma-4-E4B
+            ## Bundled
+            - `litert-lm` sidecar (offline Gemma 4 inference, opt-in via the
+              Local Models tab).
+            - LangSlice-Gemma 4 model card on Hugging Face:
+              https://huggingface.co/greenpolo/langslice-gemma-4-E4B
           releaseDraft: true
           prerelease: false
           includeUpdaterJson: false

INSTALL.md

@@ -0,0 +1,110 @@
+# LangSlice — Install (v0.1 developer preview)
+
+The v0.1 Tauri desktop app is a **developer preview**: it wraps the
+Python `langslice-harness` CLI and needs Python 3.10+ available on
+`PATH`. The estimation, registration, and quick-affine features all
+shell out to Python at runtime — the installer alone is not enough.
+
+For the no-prereq option, use the browser version at
+[greenpolo.github.io/LangSlice](https://greenpolo.github.io/LangSlice).
+The two are deliberately different products: the web demo is
+intentionally limited, the Tauri app is the feature-rich path.
+
+## Prereqs (one-time setup)
+
+### 1. Python 3.10+
+
+Any standard Windows install works. Confirm:
+
+```powershell
+python --version
+```
+
+If the command isn't found, install from
+[python.org/downloads](https://www.python.org/downloads/windows/) and
+make sure the installer's "Add python.exe to PATH" option is checked.
+
+### 2. `langslice-harness` package
+
+Clone the repo and install it in editable mode:
+
+```powershell
+git clone https://github.com/greenpolo/LangSlice.git
+cd LangSlice
+python -m venv .venv
+.venv\Scripts\activate
+pip install -e .
+```
+
+This pulls in the core deps (`brainglobe-atlasapi`, `numpy`, `scipy`,
+`pillow`, `tifffile`, `itk-elastix`, `google-genai`, …).
+
+### 3. A BrainGlobe atlas
+
+The Tauri app's first launch lets you pick + download an atlas through
+its UI, but you can pre-warm one from the command line:
+
+```powershell
+python -c "from brainglobe_atlasapi import BrainGlobeAtlas; BrainGlobeAtlas('allen_mouse_25um')"
+```
+
+This downloads to `~/.brainglobe/` and is shared across all tools.
+
+### 4. (Optional) Gemini API key
+
+For position estimation via cloud models, get a key from
+[aistudio.google.com](https://aistudio.google.com/app/apikey) and paste
+it into the app's **API Keys** tab.
+
+## Install the Tauri app
+
+1. Download `LangSlice_0.1.0_x64-setup.exe` (NSIS, ~40 MB) from the
+   [Releases page](https://github.com/greenpolo/LangSlice/releases).
+2. Double-click to install. Windows SmartScreen may warn that the
+   installer is unsigned — click **More info → Run anyway**.
+3. The app installs to `Program Files\LangSlice\`. Uninstall via the
+   standard **Add or Remove Programs**.
+4. Launch from the Start menu (`LangSlice`).
+
+The `.msi` variant (`LangSlice_0.1.0_x64_en-US.msi`) is also available
+for IT-managed deployments.
+
+## Verify it works
+
+On launch, the app's status row should show:
+
+- **Atlas**: `allen_mouse_25um` loaded (or pick another from the Atlas
+  dropdown if you downloaded one).
+- **API Keys** tab: at minimum a Gemini API key for cloud estimation,
+  or set up a local model via the **Local Models** tab.
+
+Try the **Load Demo Brain** button and run an estimate on one slice. If
+the estimate completes, the Python pipeline is wired up correctly.
+
+## Troubleshooting
+
+- **"Failed to spawn python"** — `python` isn't on PATH for the user
+  account that launched the app. Open a fresh PowerShell, run `python
+  --version`. If it works there but not from the app, log out and back
+  in (PATH changes don't always propagate to running sessions).
+- **"No module named langslice_harness"** — the `python` on PATH isn't
+  the one with `langslice-harness` installed. Either install into the
+  global Python, or update PATH so your venv's `python.exe` wins.
+- **Atlas operations hang on first run** — BrainGlobe is downloading
+  the atlas (~500 MB for `allen_mouse_25um`). Check
+  `~/.brainglobe/` for progress.
+
+## What ships in the installer
+
+- The Tauri shell + UI (~30 MB)
+- The `litert-lm` sidecar (~10 MB) + supporting DLLs for fully-offline
+  inference with the LangSlice-Gemma 4 model. The sidecar isn't auto-
+  started in v0.1; spin it up via the **Local Models** tab if you want
+  to use the local model instead of the Gemini API. Model weights are
+  not bundled — the Local Models tab links to the Hugging Face page.
+
+## Roadmap
+
+A future v0.2 will bundle Python + `langslice-harness` so the installer
+is fully self-contained and no manual setup is needed. Track progress
+on the [v0.2 milestone](https://github.com/greenpolo/LangSlice/milestones).