Tool profile

Amp

ByAmp Frontier CorporationUS

Coding agent from an independent company, with a multi-model agent, cloud machines that keep working and control via subscription or your own key.

Amp wordmark in white on a dark gradient background
Country
US
Availability
Proprietary
Modalities
  • Text
  • Code

Amp is a coding agent and development environment from Amp Frontier Corporation, an independent company created in the Sourcegraph spin-out in December 2025. The point is not just chatting in the terminal: the agent can keep working on cloud machines after you close your laptop.

What it’s for

  • Delegating multi-step coding tasks from a natural-language description
  • Running the agent locally, on the web or on dedicated remote machines
  • Keeping event-driven workflows, such as repository integrations, CI or schedules
  • Combining subagents and a meta-agent for larger tasks

How it works

  • The agent is model-agnostic and uses third-party frontier models
  • An effort selector adjusts the level of reasoning used on each task
  • You can route your own subscription or supply your own key
  • Remote machines allow execution without continuous supervision; you can also run on your own hardware

Where it runs

  • CLI/TUI for macOS, Linux and Windows
  • Web app and apps for macOS, iOS and iPad
  • Editor integrations, including VS Code, Cursor, Windsurf, Zed and Neovim
  • Cloud machines and execution on the user’s own hardware

Availability and licensing

  • The core is proprietary and there is no official source repository for the agent
  • Auxiliary components published by the company may have open licenses
  • Access is by subscription, with an option to use your own key on certain plans

How it differs

  • The current focus is persistent remote machines rather than local-only sessions
  • The company has been independent since the Sourcegraph spin-out, which changes the provider on the entry
  • It combines an interactive agent and autonomous execution in the same product

Things to note

  • Unsupervised execution requires reviewing permissions and scope
  • Consumption depends on the chosen model and the time spent running remotely
  • Features and plans change often; check the official documentation