vanilla chat interface

A simple, privacy-first web chat interface for local and cloud AI models from Ollama, LM Studio, OpenAI, Anthropic, HuggingFace, and Google, plus direct integration with the Aider, Goose, and OpenCode CLIs. Comes with 15 ice cream-themed color schemes.

quick start

starting the web interface:

npm start

Open http://localhost:2051 in your browser.

• Running Ollama/LM Studio — make sure Ollama is running (ollama serve), or start the LM Studio local server (Developer tab → Local Server → Start Server).

starting the desktop app:

Gelectron Build — lightweight with a bundled Ollama runtime:

1. Auto-install everything — ./run-this-first.sh clones and builds Gelectron + gelectron-ollama, installs dependencies, and runs the app. Options: --run (install then launch), --no-build (reuse an existing Gelectron binary).

2. Run Gelectron app binary — npm start -- -g

Electron Build — heavier with no bundled Ollama runtime:

npm start -- -e

setting up cloud models:

Pick a provider in Settings → AI Model and paste your API key.

features

• Real-time streaming — responses appear token-by-token via SSE
• Web search — toggle it in the composer or Settings; your question is searched (DuckDuckGo by default, optional Brave API key) and cited results are injected into the response
• Multi-provider — Ollama, LM Studio, OpenAI, Anthropic, HuggingFace, Google, and the Aider / Goose / OpenCode CLIs
• Compare models — run the same prompt against multiple models side by side
• Conversation management — create, switch, rename, regenerate, export (Markdown/JSON/HTML), and delete chats
• Branching edits — edit an earlier prompt without losing the original path, then move between branches
• HuggingFace installer — search GGUF models and import them into Ollama from the UI
• Custom system prompts — global default plus per-conversation override
• Auto-naming — smart chat titles generated locally via Ollama
• Themes — 15 ice cream color schemes (vanilla, strawberry, mint, lemon, lime, peach, raspberry, lavender, dragonfruit, dreamsicle, blue-moon, chocolate, monochrome, aurora, plum) plus custom JSON themes that can replace colors, accent, logo, mascot, app name, and CSS. Authoring guide: /themes/themes.html
• Settings UI — organized tabs (General, AI Model, Chat, Appearance, Shortcuts, Data & Privacy) with pickers for theme, density, text size, accent, and more
• Persistent customization — device-wide settings API plus a theme-independent custom browser/desktop icon
• Voice input — browser-native dictation
• File uploads — attach images and files with OCR/image understanding
• Native vision messages — uploaded images are translated into Ollama, Anthropic, and OpenAI-compatible multimodal formats
• Privacy — all data stays local, nothing leaves your machine unless you use a cloud provider

theme colors

These are the canonical accent, background, and text colors used by the built-in themes. Full palettes, including surfaces, borders, and message colors, live in themes/.

Theme	Accent	Background	Text
Aurora	#5CC8D7	#10151B	#EDF3F5
Blue Moon	#4F7EAE	#F3F7FA	#202830
Chocolate	#D69A6D	#1C1816	#F4ECE6
Dragonfruit	#B93F78	#FAF6F8	#30242A
Dreamsicle	#D56D38	#FBF6F1	#312720
Lavender	#7568B2	#F7F6FA	#292732
Lemon	#A98218	#FAF8F0	#2E2A20
Lime	#5F934D	#F5F8F3	#252C22
Mint	#3D876D	#F2F7F5	#222B27
Monochrome	#A1A1A6	#121214	#F4F4F5
Peach	#C66C59	#FBF5F2	#302622
Plum	#85607E	#F8F4F7	#2D272C
Raspberry	#A63F63	#FAF5F7	#30242A
Strawberry	#C84E68	#FAF5F6	#302426
Vanilla	#8A6845	#F7F5EF	#28251F

requirements

• Node.js 18+
• For local models (if using the web interface or Electron): Ollama and/or LM Studio running locally
• For CLI tools: Aider, Goose, or OpenCode installed and on your PATH
• For cloud providers: an API key from the respective service

settings

Open Settings ⌘+, — the UI is organized into sections:

• Account — your display name
• AI Model — provider and model selection, HuggingFace model installer
• Providers & API keys — enable optional providers/tools (LM Studio, Aider, Goose, OpenCode) and manage keys; LM Studio includes a connection check
• Chat — global custom system prompt
• UI — theme, density, text size, message accent, and assistant logo position
• Keyboard shortcuts / Preferences / Export / Uninstall

project structure

vanilla-sh/
├── server.js             Express server entry point
├── start.js              Launcher (web, Gelectron, Electron)
├── src/
│   ├── routes.js         API route definitions
│   ├── storage.js        File-based conversation persistence
│   ├── upload.js         File/OCR upload handling
│   ├── titles.js         Auto-naming titles
│   ├── system.js         System stats (CPU/RAM/GPU)
│   ├── providers/        Provider adapters
│   │   ├── ollama.js     Ollama
│   │   ├── lmstudio.js   LM Studio (OpenAI-compatible, local)
│   │   ├── openai.js     OpenAI + base OpenAI-compatible client
│   │   ├── anthropic.js  Anthropic Claude
│   │   ├── huggingface.js HuggingFace Inference
│   │   ├── gemini.js     Google Gemini
│   │   ├── cli.js        CLI tool base
│   │   ├── aider.js      Aider CLI
│   │   ├── goose.js      Goose CLI
│   │   └── opencode.js   OpenCode CLI
│   └── ...
├── public/
│   ├── index.html        Single-page web UI
│   ├── app.js            UI logic
│   └── styles.css
├── gelectron/            Gelectron desktop wrapper (bundled Ollama)
├── electron/             Electron desktop wrapper
├── themes/               JSON color themes + themes.html authoring guide
├── docs/                 Historical design/task notes
├── CONTRIBUTING.md       Contributor guide
├── data/
│   └── conversations/    Saved conversations (JSON)
└── config/
    └── default.json      Server configuration

api

Endpoint	Description
GET /api/health	Health check
GET /api/settings	Fetch persistent, non-secret user settings
PUT /api/settings	Update persistent user settings
GET /api/icon / PUT /api/icon	Fetch or set the theme-independent app icon
GET /api/tools	Discover model-callable tools and their JSON schemas
POST /api/tools/execute	Execute a named tool with JSON arguments
GET /api/providers	List configured providers
GET /api/models?provider=…	List available models for a provider
GET /api/system/stats	CPU/RAM/GPU usage stats
GET /api/conversations	List saved conversations
POST /api/conversations	Create a new conversation
GET /api/conversations/:id	Get conversation messages
DELETE /api/conversations/:id	Delete a conversation
POST /api/conversations/import	Import conversations from Markdown/JSON
POST /api/conversations/:id/name	Rename a conversation
POST /api/conversations/:id/regenerate	Regenerate the last response
POST /api/conversations/:id/branches/:branchId/activate	Switch to a saved conversation branch
POST /api/conversations/:id/erase-last-response	Erase the last response
POST /api/chat/stream	Send a message and stream the response (SSE); optional search / searchBackend / searchApiKey fields enable web-search injection
POST /api/chat/stop/:conversationId	Stop an in-progress stream
POST /api/chat/compare	Run a prompt against multiple models
POST /api/chat/compare-stop/:id	Stop an in-progress comparison
GET /api/search	Search conversations
GET /api/websearch?q=&backend=&key=	Web search (duckduckgo or brave) — test endpoint
GET /api/themes	List available JSON themes
GET /api/names/prewarm	Preload the auto-naming model
GET /api/hf/search	Search HuggingFace GGUF models
GET /api/hf/repo	Inspect a HuggingFace repo
POST /api/hf/install	Download a GGUF model into Ollama
POST /api/upload	Upload a file (image/audio/document)
POST /api/uninstall	Uninstall the app

configuration

Server settings live in config/default.json:

• port: server port (default 2051)
• defaultProvider: default provider (default ollama)
• providers: per-provider settings such as host, baseUrl, apiKey, defaultModel, and requestTimeout. Cloud API keys can also be entered in the app UI and are stored locally in your browser.

fonts

The fonts used in this project are subject to their respective licenses. Please consult the font files included in the repository or their upstream sources for specific license details and attribution requirements.

License

This project is currently distributed under the Diskette Labs Temporary Development License.

The source code is publicly available for transparency, learning, evaluation, testing, and authorized development. However, it may not be modified, forked, redistributed, republished, sold, sublicensed, relicensed, or used to create derivative works without express written permission from Diskette Labs.

Contributors may modify and work on the project only within the permissions granted to them through the project's contributor structure and CONTRIBUTING.md.

Diskette Labs intends to release a future version of the project under an open-source license. That future release will not automatically relicense previous versions. Versions covered by this license will remain under their existing terms and copyright unless Diskette Labs explicitly states otherwise.

See LICENSE and CONTRIBUTING.md for the complete terms and project governance.