The Amp CLI has been rebuilt from the ground up
Amp is a coding agent maximizing what's possible with today's latest AI models: autonomous reasoning, comprehensive code editing, and complex task execution. Available for free in VS Code, Cursor, Windsurf, and as a CLI.
Amp runs as a frontier coding agent that can execute multi-step tasks like running tests, fixing errors, and refactoring code across entire codebases. Users can prompt Amp to 'run tests and fix failing ones' or 'fix all TypeScript errors in this file,' and the agent handles execution autonomously with access to the full codebase context.
Amp threads allow users to save and share interactions with the agent, similar to version control for code. Teams can reference past threads, learn from each other's approaches, and continue work from previous conversations using thread URLs or IDs.
Amp automatically routes tasks to the best available models (GPT-5.5, Opus 4.7, and others) based on task complexity. Three modes—deep (extended thinking), smart (unconstrained state-of-the-art), and rush (fast)—let users get optimal performance without managing model selection themselves.
Amp is tuned to stay responsive even on the largest threads teams work with today. The agent can handle extended conversations and large file contexts without degradation, making it suitable for enterprise-scale codebases.
Amp supports plugins that hook into events, add tools, and standardize policy across a workspace. Users can write custom plugins and deploy them to their workspace, plus use AGENTS.md files to guide the agent on codebase structure, build commands, and conventions.