Agent Mini offers a 3,000-line local-first AI agent with shell, memory, and vision
Lordrovks · reddit · 2026-07-28
Agent Mini is a minimal local-first AI agent project, about 3,000 lines of Python, designed to stay readable and extensible. It ships with shell, file, web search, memory, and vision tools, uses Ollama by default, and avoids large frameworks like LangChain or LiteLLM.
The author says the goal was a practical agent that feels useful without becoming a black box. It is installable via PyPI and can be started with pip install agent-mini, agent-mini init, and agent-mini chat.
Related event: Agent Mini: A Local AI Agent in 3,000 Lines of Code(2 posts)→
More from coding & agent
- AI code review will spawn adversarial tricks, making human reviewers valuable again — bendee983 · 2026-07-28
- GlobalGPT plugs image and video generation into Codex through MCP — SarahAnnabels · 2026-07-28
- Developer Builds Auto-Updating AI Memory for SaaS Using Claude Hooks — PratikKadam_ · 2026-07-28
- Agent Skill Creator: Turn Workflows into Cross-Platform AI Skills — tom_doerr · 2026-07-28
- Reddit asks how to secure sensitive data in a multi-agent enterprise stack — Spare_Bluebird7044 · 2026-07-28
- 23-year-old ERP engineer maps a pivot from legacy software to AI application work — NelsonYus · 2026-07-28