Our goal with Amp is to let software builders harness the full power of artificial intelligence.
We believe the best way to reach that goal is to build the best possible agent we can. And we believe agents work better when they’re treated as a product end-to-end, not as a kit for users to assemble.
An agent is more than a single model. It’s a “harness” around it: a system prompt, tools, information about the environment, specialized subagents using other models — all tailored to get the most out of a specific model.
Building that harness, tuning and tweaking it, adapting it to the very best frontier models — that’s our job, not something you should have to think about. And we can do that job better if Amp isn’t used in a thousand different, mismatched ways.
What Amp has instead: agent modes. low, medium, high, and ultra. Each mode is a unique combination of models, prompts, and tools. See Models for the models that come together in a single mode.