Building my-agent: A Personal macOS Deep Agent That Lives in Your Terminal

Created on July 15, 2026

2026   ·   my-agent   ai-agents   open-source   terminal   macos   ·   ai   agents   open-source

For the past few months, I’ve been building something I’ve wanted for myself for a long time: a truly useful AI agent that runs locally on macOS, lives in the terminal, and can do real work — not just answer questions.

What is my-agent?

my-agent is a personal deep agent designed for macOS. It’s not a chatbot or a wrapper around an API — it’s a tool-using agent that can:

  • Read and edit files across your system
  • Execute shell commands and scripts
  • Browse the web for research
  • Clone repos, run builds, and deploy
  • Manage your calendar, email, and tasks
  • Remember context across sessions

All from your terminal, with full system access, running against any LLM provider you choose.

Why I built it

Existing AI coding assistants (Copilot, Cursor, Claude Code) are powerful but limited. They work inside specific editors, have narrow scopes, and can’t manage the full lifecycle of a project — let alone coordinate across multiple repos, services, and tools.

I wanted something that could:

  1. Work anywhere — terminal, IDE, messaging platforms
  2. Maintain persistent context — not just within a session, but across weeks
  3. Self-improve — learn from past work and save reusable patterns
  4. Run locally first — no cloud dependency for core operations

Key design decisions

Provider-agnostic from day one

my-agent doesn’t lock you into one LLM. It works with OpenAI, Anthropic, Google, DeepSeek, xAI, OpenRouter, and local models via llama.cpp or Ollama. You can swap providers mid-conversation.

Skills — the agent that learns

Every time my-agent solves a complex problem or discovers a workflow, it can save that as a skill — a reusable markdown document with instructions, pitfalls, and verification steps. Over time, it accumulates domain knowledge specific to your work.

Persistent memory

Unlike most agents that reset every session, my-agent has cross-session memory. It remembers your preferences, project conventions, and past decisions. This means it gets better the more you use it.

Multi-platform gateway

The same agent runs on Telegram, Discord, Slack, WhatsApp, and a dozen other platforms. You can start a task on your terminal, check progress from your phone, and review results on any device.

What’s next

  • Voice integration — speak to your agent, get voice responses
  • Multi-agent orchestration — spawn worker agents for parallel tasks
  • Deeper IDE integration — ACP server for VS Code, Zed, JetBrains
  • Mobile companion — native iOS/Android app

The repo is open source at github.com/vikasudasi/my-agent. If you’re building in the AI agent space, I’d love to collaborate — reach out on X or LinkedIn.

This is part of a broader effort to make AI agents genuinely useful, not just impressive demos. More to come.