Amp is the agentic coding tool from Sourcegraph, available as both a terminal CLI and a VS Code extension. It builds on Sourcegraph’s decade of experience in code search and intelligence, giving the underlying model precise, semantically relevant context from large repositories rather than relying on raw token stuffing.
Key capabilities
Codebase search integration — Amp uses Sourcegraph’s code search infrastructure to retrieve the most relevant symbols, call sites, and files before sending context to the model. On large monorepos where context-window limits matter, this retrieval step meaningfully improves answer quality.
File system and shell access — The agent reads, creates, edits, and deletes files and runs terminal commands (build tools, test runners, git). Changes are proposed as diffs and applied on developer approval.
VS Code and terminal dual delivery — Amp ships as both a VS Code extension and a standalone CLI, so developers can use it in whichever environment fits their workflow.
Git integration — Creates branches, stages changes, writes commit messages, and opens pull requests.
Multi-model support — Supports Anthropic Claude and OpenAI GPT-4o. Model choice is configurable per session.
Autonomy level
Level 3 (supervised agent): each file edit and shell command is proposed and applied only after developer approval. The agent can chain multiple edits in a single prompt but shows a summary checkpoint before committing.
Strengths
- Sourcegraph’s codebase search gives better retrieval on large repos than naive context stuffing
- Dual CLI + VS Code delivery fits both terminal and editor workflows
- Clean diff-based approval flow
- Strong on existing large codebases where the model needs to find the right call site
Limitations
- Closed source; pricing model subject to change
- Retrieval quality depends on repository structure and Sourcegraph’s indexing
- Smaller user base and community than Cursor or Claude Code