Ampcode is a coding agent and development environment for frontier development. It is built on Sourcegraph's code search infrastructure, has a deep codebase graph, and does not limit token usage.
In terms of purpose, Ampcode is designed to help developers build software more efficiently, reduce the burden of managing multiple agents, and let agents continue working in the background, thereby improving delivery efficiency.
Main capabilities include: agents can run continuously, continuing to work even after the laptop is closed; users can continue interacting via phone, or let agents continue executing tasks while sleeping; agents can wake themselves up to check the production environment, for example monitoring performance releases; support for sharing agents with teammates; each agent can have its own orb, allowing it to run anywhere; users can grant orb production environment access permissions from the terminal (such as via OIDC and Google Cloud Grants); agents can access production logs and analyze them; support for describing problems via video (such as screen recording flicker) so the agent can locate and fix them; support for running Storybook in Portal for frontend testing; support for one-click publishing of changes via the Changes tab, and instructing the agent to run tests, rebase to origin main, and publish to main; after publishing, the agent can monitor release status at specified intervals (such as every 15 minutes) and notify the user once it is live.
In terms of usage, users can run agents locally on their machines via CLI and runner, or run them in the cloud via orb. In the workflow, after users send a prompt, they can close the laptop, and the agent will continue working. Agents can wake themselves up, check production metrics, generate images, summarize fixes, create Storybook, publish changes, and more. Users can also use dictation to input prompts and approve orb production access via the amp threads continue command.
Applicable scenarios include: performance monitoring and releases in daily development work, bug fixing, frontend UI issue debugging, feature promotion (such as promoting an experimental feature from a feature flag to a user-enableable experimental feature), production environment monitoring and alerting, and tasks that need to run for a long time.
In terms of limitations, the official materials do not explicitly mention specific limiting conditions, such as whether it is completely free, whether there are usage quotas, whether it supports all programming languages, etc. Some features (such as the experimental feature Recap) may generate additional inference requests and costs, but specific cost details are not yet provided. In addition, information about platform support, system requirements, etc. is also not yet provided.
