Amp

Coding agent and development environment that runs autonomous agents in cloud orbs or locally via CLI.

Overview

Amp is a coding agent and development environment designed for autonomous software development. It lets you send a prompt and close your laptop while agents continue working in isolated cloud environments called "orbs," or run locally via CLI. The platform supports multiple AI models (GPT-5.6, Claude, Fable 5.1) and provides a unified experience across web, macOS, iOS, and terminal interfaces.

Key to Amp's design is the concept of orbs—ephemeral machines created per thread with your code, tools, and plugins pre-configured. You can start an orb from your iPhone, check on it from your Mac, and follow up in the terminal. Agents can be granted production access via OIDC, run automations on schedules, and integrate with external services through MCP and plugins.

Amp is built by an independent agent research lab that prioritizes frontier capabilities over backward compatibility, meaning the product evolves rapidly with new AI model capabilities.

Key features

Orbs: cloud machines per thread that persist after you close your laptop
Multi-model support: GPT-5.6, Claude, Fable 5.1
Cross-platform: web, macOS, iOS, CLI, and terminal
Production access via OIDC for agents
Automations and event-driven orbs
Portals for testing and sharing
Slack integration via MCP
Multiplayer and team workspaces
Bring your own keys (BYOK) and model routing

Pros

Agents continue working after you close your laptop
Unified experience across web, desktop, mobile, and CLI
Fine-grained production access control for agents
Multi-model routing optimizes for each model's strengths
Rapid iteration on frontier AI capabilities
Strong team collaboration features

Cons

No backward compatibility; product evolves rapidly
Requires learning new concepts like orbs and portals
Pricing scales with orb usage minutes
Not designed for users seeking stable, unchanging workflows

Use this if

You need autonomous agents that can work unsupervised in cloud environments, want to collaborate with teammates on agent-driven development, or need fine-grained control over agent access to production systems.

Skip this if

You require stable, unchanging software with backward compatibility guarantees, or prefer traditional development workflows without autonomous agents.

Best for

Autonomous code generation and bug fixesRunning long-lived agent tasks without supervisionMulti-model AI workflowsTeams collaborating on agent-driven developmentDevelopers working across multiple devices