LocalLLMMind

A production-grade, privacy-first desktop AI workstation for local LLMs.
Engineered by Kapil Kumar Yadav

🌟 Overview

LocalLLMMind is a full-featured desktop AI workstation that interfaces directly with Ollama to run LLMs 100% locally. Zero data leaves your machine. It combines the polish of modern AI products with hardware-optimized streaming, live benchmarking, reasoning model introspection, multimodal vision, RAG document chat, and a complete model management toolkit.

✨ Features

👁️ Multimodal Vision Support

Paste, drag-and-drop, or browse images directly into the chat composer.
Automatic Base64 conversion to Ollama's /api/chat images array.
Click-to-zoom lightbox for high-resolution image inspection.
Supports llama3.2-vision, llava, moondream, and all Ollama vision models.

⚡ Hardware-Optimized Streaming (30–80+ tok/s)

Tokens are buffered on a 35ms micro-interval (~30fps), reducing React re-renders by ~75%.
Jitter-free auto-scroll during active generation.
Live generation metrics badge: speed (⚡ 54.2 tok/s), token count, duration, and active model.

🧠 DeepSeek-R1 & Reasoning Model Support

Built-in <think>...</think> tag parser for DeepSeek-R1, Qwen 2.5, and reasoning models.
Interactive collapsible thought accordion with live pulsing indicator during chain-of-thought.

📄 RAG / Document Chat ("Chat with Docs")

Drag-and-drop or browse 40+ file formats: source code (.py, .js, .ts, .rs, .go, .java, .cpp), configs (.json, .yaml, .toml, .env), data (.csv, .sql), and docs (.md, .txt).
Client-side text extraction and vector chunking sandbox — nothing leaves your machine.
500KB safety guard to prevent context window overflows.

📥 In-App Model Manager

View installed models with disk size, parameter count, quantization format, and model family.
Pull new models with real-time progress bars and cancel support.
1-click quick-pull chips for popular models (llama3.2, deepseek-r1:8b, phi4, gemma3).

✏️ Message Edit & Regenerate

Edit any user message — truncates the conversation and streams a fresh response from that point.
Regenerate assistant responses with full version history.
Version navigation — browse between regenerated responses with ◀ 2/3 ▶ arrows.

🎭 AI Personas

6 built-in system personas: General Assistant, Software Architect, Bug Hunter, Concise & Direct, Academic Researcher, Creative Writer.
Switch personas per-conversation from the chat header.

⚔️ Arena Mode (Side-by-Side Comparison)

Run two models simultaneously on the same prompt.
Compare responses side-by-side with live generation speeds (tok/s) and vote for the better output.

📝 Prompt Library & Slash Commands

Built-in prompt template library with CRUD, categories, and import/export.
Type / in the composer for instant slash-command autocomplete.

🔍 Global Conversation Search (`Cmd+Shift+F`)

Full-text search across all conversations with highlighted match snippets.
Results grouped by conversation with 1-click jump to the exact message.

🧠 Smart Auto-Title Generation

After the first exchange, the LLM automatically generates a descriptive 3–5 word title for the conversation.

📁 Project Folders

Create color-coded project folders to organize conversations.
Collapsible sidebar tree with chat count badges.
Move conversations between projects.

🎙️ Voice Input & Text-to-Speech

Speak prompts using Web Speech Recognition with live interim feedback.
Listen to assistant responses with markdown-sanitized TTS output.

📊 Context Window Meter

Live header chip showing estimated tokens vs. context limit (e.g. ~1.2k / 4k tok).
Color-coded thresholds: blue → amber (65%+) → red (85%+).

💾 Local Storage & Disk Sync

Choose between browser memory or a local directory on your hard drive.
Auto-sync conversations to disk as JSON + readable Markdown files (Obsidian-compatible).
Full JSON backup import/export with data portability.

📤 Share & Export

Export as PNG image, Markdown, plain text, or PDF (print-optimized A4).
Native OS Web Share API integration.

🌐 Web Search Grounding

Ground LLM responses with real-time web search context.
Source citations displayed as clickable chips on messages.

🧪 Artifact Sandbox

Interactive sandbox drawer for rendering HTML/CSS/JS artifacts generated by the LLM.

⌨️ Custom Keyboard Shortcuts

Full shortcuts manager with conflict detection and custom keybinding creation.
Bind custom key combos to prompt templates and quick actions.

Shortcut	Action
`Cmd+K`	New Conversation
`Cmd+B`	Toggle Sidebar
`Cmd+F`	Find in Chat
`Cmd+Shift+F`	Global Search
`Cmd+Shift+M`	Model Manager
`Cmd+,`	Settings
`Cmd+/`	Shortcuts Manager
`Escape`	Stop Generation / Dismiss

📱 Progressive Web App

Installable as a standalone desktop app via Chrome/Edge.
Offline caching service worker for app shell and static assets.

🎨 Custom UI Themes & Appearance Editor

8 Curated Designer Presets: Modern Dark, Clean Light, Midnight Obsidian, Cyberpunk Neon, Tokyo Night, Forest Emerald, Sunset Glow, and Royal Amethyst.
Custom Accent Color Swatches: Pick from curated neon/pastel accents or custom hex values with instant live preview.
Message Bubble Roundness: Granular slider from sharp (4px) to pill (24px) bubble shapes.
Typography Scale Adjuster: Adjust font scaling (Small, Medium, Large, Extra Large) across all chat and workspace interfaces.
Persistent Preferences: Theme configuration syncs across restarts via local storage.

⚡ Autonomous Agent Mode & Implicit Live Tools

Implicit Live Web Search: The agent queries DuckDuckGo for live facts, current documentation, and news without requiring manual approval clicks, citing sources directly.
Real-Time Weather Integration: Live weather conditions and forecasts via Open-Meteo.
Interactive Agent Activity Log: View step-by-step tool traces, execution times, token counts, and export trace JSONs.
Collapsible Activity Feed: Quick "Hide activity" / "Show activity" toggle with persistent visibility preference.
Human-in-the-Loop Safe Execution: Destructive actions (modifying files and executing shell commands) require explicit approval.

🌐 Multi-Provider AI Support (Local & Cloud)

First-Class Ollama Integration: 100% private, native local execution.
Top Cloud Providers: Seamlessly toggle between OpenAI, Anthropic Claude, Google Gemini, and xAI Grok.
TypeSafe Jev Integration: Native adapter for deterministic and structured LLM decision workflows.
Custom Endpoints: Connect any OpenAI-compatible API gateway with custom base URLs and headers.

📐 TypeSafe Jev Decision Studio

Deterministic Evaluation Workflows: Model selection, structured state parameters, and decision criteria matrix.
Strict Output Validation: Ensure schema compliance and structured JSON guarantees.
Live Test Suite: Execute multi-attribute scoring and benchmark decision matrices before production deployment.

🌓 Dark & Light Theme

Curated glassmorphism design with fluid gradients and subtle micro-animations.
Fully responsive — works on desktop, tablet, and mobile.

🚀 Quickstart

Prerequisites

Install Node.js (v18+).
Install and run Ollama:
```
ollama serve
```

Pull a model:

ollama pull llama3.2
# For vision:
ollama pull llama3.2-vision:11b

Installation

git clone https://github.com/kapilyadav22/local_llm_ui.git
cd local_llm_ui
npm install
npm run dev

Open http://localhost:2210/ in your browser.

Production Build

npm run build
npm run preview

🐳 Docker Deployment

⚡ Quick Start via Docker Hub (No build required)

Pull and run the pre-built, lightweight Alpine image directly from Docker Hub:

docker run -d \
  --name localllmmind \
  -p 3000:80 \
  --add-host=host.docker.internal:host-gateway \
  -e OLLAMA_URL=http://host.docker.internal:11434 \
  --restart unless-stopped \
  kapilyadav22/localllmmind:latest

Open http://localhost:3000 in your browser.

📦 Docker Hub Repositories:

kapilyadav22/localllmmind (Official)

kapilyadav22/runlocalllm (Mirror)

Docker Compose (Recommended)

# Start LocalLLMMind → http://localhost:3000
docker compose up -d

Automatically connects to host Ollama at http://host.docker.internal:11434.

All-in-One (LocalLLMMind + Ollama containerized):

docker compose --profile with-ollama up -d

Build from Source

docker build -t localllmmind .
docker run -d \
  -p 3000:80 \
  --add-host=host.docker.internal:host-gateway \
  -e OLLAMA_URL=http://host.docker.internal:11434 \
  --name localllmmind \
  localllmmind

Variable	Default	Description
`PORT`	`80`	Nginx listening port inside container
`OLLAMA_URL`	`http://host.docker.internal:11434`	Ollama API endpoint on host gateway

🏗️ Project Structure

localllmmind/
├── src/
│   ├── components/
│   │   ├── Chat/           # ChatView, MessageBubble, MessageInput, WelcomeScreen,
│   │   │                   # ArenaMessageBubble, ContextMeter, PersonaDialog,
│   │   │                   # PromptLibraryDialog, ShareChatDialog, ArtifactSandboxDrawer
│   │   ├── Layout/         # AppLayout, Sidebar, GlobalSearchDialog, ProjectDialog
│   │   ├── Settings/       # SettingsDialog, ModelManagerDialog
│   │   └── common/         # MarkdownRenderer, ShortcutsManager, AppLogo, ModelSelector
│   ├── constants/          # App branding, models, personas, shortcuts
│   ├── hooks/              # useAudio (Speech-to-Text & TTS)
│   ├── services/           # ollamaService, webSearchService
│   ├── store/              # React Context + useReducer state management
│   ├── utils/              # Storage, documentUtils, pdfExport, dialogService
│   ├── theme.js            # MUI Dark & Light theme tokens
│   └── main.jsx            # Entrypoint with PWA service worker registration
├── public/                 # PWA manifest, service worker, icons
├── Dockerfile              # Multi-stage Node + Nginx Alpine build
├── docker-compose.yml      # Production orchestration with Ollama profiles
├── nginx.conf.template     # Streaming reverse proxy for LLM tokens
└── vite.config.js          # Vite config with Ollama API proxy

Code workspace: opening projects and adding context

Antigravity-Style Agent Chat: Pinned bottom composer for continuous follow-ups, auto-scrolling tool streams, and 1-click file proposal application.
GitHub Sync & Gists: Link your GitHub account via Personal Access Token. Push commits to existing or new repositories, or export code to Gists.
Project Import & Folders: Connect local folders directly or import ZIP archives (preserves relative paths, auto-skips node_modules and binaries).
Persistent Terminal & Runner: Full interactive terminal sessions with persistent state across reloads, command history, and background processes.
Code Formatting & Git: Syntax-aware formatting (Prettier & Black), file staging, line review comments, and unified diff inspection.

Agent workspace, model controls, and evaluations

Model Manager: Inspect, warm, or unload active models in Ollama memory without deleting downloaded weights.
Autonomous Agent Mode: Tool-calling agent with implicit web search, document retrieval, and collapsible step-by-step activity traces.
Code Agent Workspace: Autonomous repository scanning, file diagnosis, refactoring, and code proposal generation.
Agent Traces & Activity: Real-time duration metrics, token counters, and downloadable JSON execution traces.
Playground & Evaluations: Custom prompt tuning, parameter presets, and automated fixture evaluations with assertion scoring.

🎬 Product Demo Video & Screenshots

Interactive Walkthrough Video: docs/assets/demo_video.mp4 (Full desktop walkthrough: Developer Chat, Autonomous Agent Mode, GitHub linking & remote sync, Theme Studio, and Multimodal Vision)

🖼️ Real Feature Scenarios & High-Resolution Screenshots

#	Scenario / Feature	Real Screenshot	Description
1	Antigravity IDE Agent Chat	Inspect Preview	Multi-turn developer agent chat with continuous auto-scrolling, code proposals, and sticky bottom prompt composer.
2	GitHub Account Linking	Inspect Preview	Git Source Control sidebar with GitHub PAT token connection, account badge, and remote repository sync options.
3	Push Project to GitHub	Inspect Preview	Connected GitHub account profile with 1-click Push to remote repository, branch target, and commit authoring.
4	Multimodal Vision	Inspect Preview	Upload images with real-time thumbnail preview, size badge, and visual layout/color analysis prompt.
5	Code Studio Refactor	Inspect Preview	Full project file explorer with Python/JS syntax highlighting and AI Assistant refactoring prompt.
6	Local RAG & Knowledge Base	Inspect Preview	Local RAG document dialog showing chunking statistics, token counts, and semantic retrieval sandbox.
7	Extensive Code Review & Comments	Inspect Preview	Multi-tab code workspace with file tree, editor, and dedicated line-targeted review comments panel.
8	AI Providers Configuration	Inspect Preview	Unified provider settings: OpenAI, Anthropic Claude, Google Gemini, xAI Grok, Jev, Ollama, and Custom endpoints.
9	TypeSafe Jev Decision Studio	Inspect Preview	Deterministic decision studio modal with model selector, state schema editor, and evaluation criteria matrix.
10	Model Comparison & Arena	Inspect Preview	Side-by-side responses (Llama 3.2 vs DeepSeek-R1) with live `tok/s` benchmarks, token counts, and voting chips.
11	Chat Dashboard	Inspect Preview	Full workstation home showing sidebar, project folders, model selector, context meter, and quick prompts.
12	Autonomous Agent Mode (Active)	Inspect Preview	Step-by-step tool trace, live implicit DuckDuckGo search execution, and collapsible activity panel.
13	Autonomous Agent Mode (Clean)	Inspect Preview	Streamlined view with activity collapsed for distraction-free reading of grounded agent answers.
14	Theme Customizer & Swatches	Inspect Preview	8 curated theme presets, custom accent color swatches, bubble roundness slider, and font scale tuner.
15	Evaluations & Benchmark Lab	Inspect Preview	System prompt tuning, fixture test runs, pass/fail assertion metrics, and model latency comparisons.

👨‍💻 Author

Kapil Kumar Yadav — Senior Software Engineer

💖 Sponsor & Support

If you find LocalLLMMind valuable for your workflow, consider sponsoring or supporting this project:

💖 Sponsor on GitHub Sponsors
⭐ Star this repository to help others discover it
🐛 Report bugs or submit features to help improve the project

📄 License

Licensed under the Apache License 2.0.

Name		Name	Last commit message	Last commit date
.github		.github
docs		docs
public		public
scripts		scripts
server		server
src		src
tests		tests
.dockerignore		.dockerignore
.gitignore		.gitignore
.oxlint_out.txt		.oxlint_out.txt
.oxlintrc.json		.oxlintrc.json
DOCKER_HUB_README.md		DOCKER_HUB_README.md
Dockerfile		Dockerfile
LICENSE		LICENSE
README.md		README.md
docker-compose.yml		docker-compose.yml
index.html		index.html
nginx.conf.template		nginx.conf.template
package-lock.json		package-lock.json
package.json		package.json
tsconfig.json		tsconfig.json
vite.config.js		vite.config.js

Uh oh!

Latest commit

History

Folders and files

Repository files navigation