Neo by Amp.

Neo by Amp

Neo is the completely rebuilt Amp CLI, designed for the next generation of AI coding agents. Built on a remote-controllable, compaction-first, plugin-powered architecture, it eliminates manual context management and delivers dramatically improved performance for developers who want agents with longer leashes and less handholding.

Product Highlights

• Automatic Context Compaction: When your context window fills to 90%, Amp automatically summarizes and compacts threads so you never have to manually manage context or panic about running out of space.
• Remote Control from Anywhere: Start a thread in your terminal and remotely control it from ampcode.com with live updates, message queuing, and real-time intervention capabilities.
• Plugin-Powered Extensibility: The new Amp Plugin API lets you handle events, register custom tools, add commands, show UI elements, and even ask AI questions with confidence scoring—all through simple TypeScript files.
• 79% Less CPU, 70% Less Memory: Neo runs dramatically more efficiently than the previous CLI, handling 5000+ message threads without slowdowns.
• Smart Message Queuing: Messages queue by default when the agent is busy, with steering controls to fast-track critical inputs without interrupting ongoing work.

Use Cases

• Long-Running Development Sessions: Let agents work autonomously for hours while you monitor and intervene remotely from any device, perfect for complex refactoring or multi-file implementations.
• Custom Agent Workflows: Build organization-specific permissions, custom tools, and specialized commands using the Plugin API to match your team's security policies and development practices.
• High-Volume Code Operations: Handle massive threads with thousands of messages without performance degradation, ideal for large-scale migrations or extensive codebase analysis.

Target Audience

Neo is built for professional developers and engineering teams who rely on AI coding agents for substantial development work and need a system that scales with increasingly capable frontier models. It's ideal for those frustrated by context limits, manual handoffs, and the babysitting required by older agent interfaces.