MacVis — Local-First AI Desktop Assistant
A fully-agentic macOS desktop app that gives AI full access to your Mac — bash, filesystem, web search, and MCP integrations. Bring-your-own-key: Anthropic, OpenAI, OpenRouter, Gemini, Groq, and Ollama. No backend, no telemetry, everything stays on your machine.
The Challenge
Cloud agent platforms like ChatGPT and Claude Desktop are powerful but keep all data on remote servers, charge per-message, and can't access local files or run arbitrary Mac commands. Developers needed a local-first alternative with genuine system access.
Approach & Solution
Built a native macOS Electron app with a React + TypeScript frontend. The main process owns all I/O, AI calls, and tool execution — the renderer is pure UI communicating over a typed IPC bridge. Implemented live API key validation that hits each provider's /models endpoint, persistent chat history stored as flat JSON, and a Projects view that auto-discovers code the agent creates.
Results & Impact
- Full bash/filesystem/web-search access out of the box with zero config
- Live key validation across 6 providers — Anthropic, OpenAI, OpenRouter, Gemini, Groq, Ollama
- Persistent memory: every session saved to ~/.macvis/sessions/ and reloaded on launch
- Projects view auto-discovers and runs agent-created code (HTML, Node, Next.js, Python)
- MCP-ready with first-class config for GitHub, Supabase, Vercel, Railway, Slack
- 100% local-first — no analytics, no telemetry, no remote logging
More Projects
Claude-Powered iOS AI Assistant
Full-stack iOS AI assistant with natural-language control over device functions, Gmail/Calendar automation, AI-participated group chat, and hybrid pgvector + Supabase persistent memory system.
Full StackAI Trading Strategy Generator
End-to-end agentic backend generating 200+ trading strategy variants autonomously. Client deployed 3 profitable strategies to production with measurable revenue impact.
Full StackBrand Performance SaaS Platform
AI-powered platform integrating 5+ LLMs with automated evaluation pipelines, cutting manual brand auditing effort by 90% and directly improving customer outcomes.