Add Dream Server desktop dev source

.gitignore

@@ -4,6 +4,8 @@ __pycache__/
 *.egg-info/
 dist/
 build/
+!resources/dev/dream-server-desktop/build/
+!resources/dev/dream-server-desktop/build/**
 
 # Logs
 *.log

.gitleaks.toml

@@ -16,4 +16,5 @@
     '''dream-server/installers/phases/06-directories\.sh''',
     '''dream-server/installers/macos/lib/env-generator\.sh''',
     '''dream-server/installers/windows/lib/env-generator\.ps1''',
+    '''resources/dev/dream-server-desktop/vendor/hermes-agent/tests/hermes_cli/test_plugins_cmd\.py''',
   ]

resources/dev/README.md

@@ -45,6 +45,16 @@ A lightweight Docker image containing everything needed to run the DreamServer i
 
 Static HTML download page for `dreamserver.dev/download` (or equivalent). Auto-detects visitor's OS and shows the right installation command. Single-file, no dependencies, deploys anywhere.
 
+### [`dream-server-desktop/`](dream-server-desktop/) — Electron Desktop Shell
+
+Source-first Electron desktop build for Dream Server DESKTOP. It includes the
+renderer, shared local runtime, Hermes Agent vendor source, CLI/gRPC entrypoints,
+and cross-platform packaging scripts for Windows, Linux, and macOS.
+
+Generated artifacts stay out of Git: `node_modules`, Electron `dist/` output,
+Python virtual environments, downloaded standalone Python runtimes, GGUF models,
+and machine-specific llama.cpp binaries.
+
 ### [`extensions-library/`](extensions-library/) — 33 Service Extensions
 
 **The next wave of DreamServer services.** 17 extensions are already in production — these 33 are being tested for the next release.

resources/dev/dream-server-desktop/.gitattributes

@@ -0,0 +1,24 @@
+* text=auto eol=lf
+*.bat text eol=crlf
+*.cmd text eol=crlf
+*.ps1 text eol=crlf
+*.png binary
+*.jpg binary
+*.jpeg binary
+*.gif binary
+*.ico binary
+*.icns binary
+*.mp4 binary
+*.wav binary
+*.ttf binary
+*.woff binary
+*.woff2 binary
+*.asar binary
+*.exe binary
+*.dll binary
+*.pak binary
+*.dat binary
+*.bin binary
+bin/llama/** filter=lfs diff=lfs merge=lfs -text
+src/shaders/** whitespace=-blank-at-eol,-blank-at-eof
+vendor/** whitespace=-blank-at-eol,-blank-at-eof

resources/dev/dream-server-desktop/.gitignore

@@ -0,0 +1,51 @@
+node_modules/
+
+# Electron/build outputs
+dist/
+dist-*/
+out/
+release/
+.asar-rebuild/
+.asar-rebuild-out/
+.asar-rebuild-out-*/
+builder-debug.yml
+*.blockmap
+
+# Local runtime binaries/models
+resources/python/
+bin/llama/
+/models/
+*.gguf
+
+# Local state and logs
+.tmp-preview-control.json
+*.log
+npm-debug.log*
+yarn-debug.log*
+yarn-error.log*
+pnpm-debug.log*
+
+# Legacy/extracted root copies that are not used by package.json
+/core.js
+/main.js
+/renderer.js
+/index.html
+/styles.css
+/rebuild-shader.cjs
+/*-screen.png
+
+# Python/cache artifacts
+**/__pycache__/
+*.py[cod]
+*.pyo
+.pytest_cache/
+.ruff_cache/
+
+# Local environment files
+.env
+.env.*
+!.env.example
+
+# Vendored source: keep full local vendor snapshots.
+vendor/**/.git/
+.vendor-git-backup/

resources/dev/dream-server-desktop/README.md

@@ -0,0 +1,125 @@
+# Dream Server DESKTOP
+
+Dream Server DESKTOP is the Electron desktop source tree for the local-first
+Dream Server shell. It packages the renderer, Node runtime, Hermes Agent vendor
+source, CLI, gRPC server, and build scripts needed to create installable desktop
+artifacts.
+
+This directory is source-first by design. It does not commit `node_modules`,
+compiled Electron output, Python virtual environments, GGUF models, or
+platform-specific llama.cpp binaries.
+
+## What is included
+
+- Electron shell: `app-main.js`, `preload.js`, `src/`
+- Shared runtime: `runtime/`
+- CLI entrypoint: `bin/dream.js`
+- gRPC entrypoint: `runtime/grpc-server.js`
+- Hermes Agent vendor source: `vendor/hermes-agent/`
+- Browser harness vendor source used by Hermes: `vendor/browser-harness-upstream/`
+- Desktop build scripts: `scripts/`
+
+## Requirements
+
+- Node.js 20 or newer
+- npm
+- `tar` available on PATH
+- Internet access when downloading standalone Python runtimes
+
+For a fully local LLM, run an OpenAI-compatible endpoint such as llama.cpp,
+Ollama, LM Studio, or the existing Dream Server local model server, then point
+the app at that base URL and model in Settings. Models and llama.cpp builds are
+machine-specific, so they are intentionally not versioned here.
+
+## Development
+
+```bash
+npm install
+npm start
+```
+
+The first Electron launch prepares Hermes Agent if the `.venv-hermes` virtual
+environment is missing. You can also run the setup explicitly:
+
+```bash
+npm run setup:hermes
+```
+
+Useful overrides:
+
+- `DREAM_HERMES_PYTHON`: absolute path to a Python interpreter
+- `DREAM_HERMES_VENV_DIR`: absolute path for the Hermes virtual environment
+- `DREAM_APP_USER_DATA`: test-only user-data directory override
+- `HERMES_GIT_BASH_PATH`: Git Bash path on Windows when local shell tools need it
+
+## Tests
+
+```bash
+npm run test:runtime
+```
+
+## Standalone Python Runtime
+
+The desktop builds use python-build-standalone archives and extract them into
+`resources/python/<platform>-<arch>/`. These files are generated locally and are
+ignored by Git.
+
+Download the runtime for the current platform:
+
+```bash
+npm run download:python
+```
+
+Download explicit targets:
+
+```bash
+npm run download:python:win
+npm run download:python:linux
+npm run download:python:mac
+npm run download:python:all
+```
+
+The downloader verifies that the standard library contains `encodings/cp437.py`,
+`zipfile.py`, `venv`, and `ensurepip`. On the current host platform it also runs
+a Python smoke test. This prevents packaged builds from shipping an incomplete
+Python runtime that would later fail in `pip` with errors such as
+`LookupError: unknown encoding: cp437`.
+
+## Packaging
+
+Build the unpacked app for the current platform:
+
+```bash
+npm run pack
+```
+
+Build installable artifacts:
+
+```bash
+npm run dist:win
+npm run dist:linux
+npm run dist:mac
+npm run dist:mac-arm64
+npm run dist:mac-x64
+```
+
+Expected artifacts:
+
+- Windows: portable `.exe`
+- Linux: AppImage and `.deb`
+- macOS: `.dmg` and `.zip`
+
+`npm run dist:win`, `npm run dist:linux`, and the macOS dist commands download
+the matching standalone Python runtime before invoking `electron-builder`.
+
+## Local Models and llama.cpp
+
+This source tree does not include a bundled GGUF model or compiled llama.cpp
+server. To use a local model, either:
+
+- configure an OpenAI-compatible endpoint in the desktop Settings screen, or
+- place a compatible `llama-server` build under `bin/llama/<platform>-<arch>/`.
+
+The runtime resolver checks canonical platform directories such as
+`win32-x64`, `linux-x64`, `darwin-arm64`, and `darwin-x64`, with compatibility
+aliases for older local layouts.

resources/dev/dream-server-desktop/THIRD_PARTY_NOTICES.md

@@ -0,0 +1,29 @@
+# Third Party Notices
+
+This experimental Dream Server Hermes branch vendors Hermes Agent as an
+architecture and implementation reference under `vendor/hermes-agent`.
+
+Hermes Agent is distributed under the MIT License:
+
+Copyright (c) 2025 Nous Research
+
+The full license text is available at:
+
+- `vendor/hermes-agent/LICENSE`
+
+This branch uses Hermes through a stdio bridge under `runtime/hermes`, which
+launches the vendored Python runtime and imports Hermes' own `AIAgent` rather
+than reimplementing its agent loop inside the Electron renderer.
+
+The code workbench shader controls are inspired by Ghostty shader projects:
+
+- `vendor/ghostty-shader-playground` / https://github.com/KroneCorylus/ghostty-shader-playground
+- https://github.com/sahaj-b/ghostty-cursor-shaders
+
+Selected GLSL shader files from those MIT-licensed projects are bundled under
+`src/shaders/ghostty` and are wrapped at runtime for the Electron code
+workbench's WebGL renderer.
+
+The code workbench also bundles JetBrains Mono Regular under `src/fonts` for
+Ghostty-like terminal typography. JetBrains Mono is distributed under the SIL
+Open Font License; the copied license is `src/fonts/OFL-JetBrainsMono.txt`.