Amp (Sourcegraph)

✓ Editorially verified

Frontier coding agent from Sourcegraph with pass-through model pricing

Freemium· Pay-as-you-go with $5 minimum credit purchase, zero markup on provider API costs; Enterprise tier from $1,000 USD minimum with SSO/SAML, zero data retention, cost controls, and MCP allowlisting.CodingGPT-5.5, Claude Opus 4.8, GPT Image 2 (frontier multi-model)8.4 / 10

Visit website →

Best for

Professional developers and engineering teams who want unconstrained access to frontier coding models, agentic workflows, and remote background execution without subscription markups.

Skip if

Hobbyists on tight budgets, teams needing predictable flat-rate billing, or developers who just want inline autocomplete rather than agentic tool use.

Amp is Sourcegraph's frontier coding agent designed for developers who want unconstrained access to top-tier models like GPT-5.5 and Claude Opus without paying subscription markups. It runs from the terminal, integrates with major IDEs, and can be driven remotely from a web or mobile client via passkey-authenticated sessions. Unlike traditional pair-programming assistants, Amp is built around agentic workflows: you can spin up subagents for parallel tasks, hand off deep reasoning to an Oracle model, search across public and private GitHub repositories with a Librarian tool, generate and edit images inline with Painter, and run automated code reviews with customizable rules. Three usage modes give you cost-vs-quality control: Deep for extended reasoning on hard problems, Smart for state-of-the-art models with no token constraints, and Rush for quick, well-defined tasks. Amp supports AGENTS.md files for per-project instructions, a Skills system for repeatable task templates, and MCP (Model Context Protocol) for wiring in custom tools and data sources. IDE integrations cover VS Code, Cursor, JetBrains (2025.1+), Neovim, and Zed, all of which share awareness of the currently open file and selection with the CLI. Remote 'orb' execution lets long-running agents work in the background while you monitor from your phone. It's aimed at professional developers, tech leads, and engineering teams who already understand how to prompt agents effectively and want a tool that stays on the model frontier rather than optimizing for cheap tokens.

Editor's take

Amp is one of the more honest offerings in the coding-agent space: Sourcegraph passes through model costs at zero markup and leans into the frontier rather than pretending a small model is enough. If you're already comfortable directing agents and want Oracle-grade reasoning plus remote orbs, it's worth the credit minimum. Just watch your spend.

— The AI Tool Bible editorial team

Pros

✅ Zero-markup pass-through pricing on top-tier models keeps costs honest and predictable
✅ Access to frontier models (GPT-5.5, Claude Opus 4.8) without artificial token caps
✅ Rich agentic toolkit: Oracle reasoning, Librarian code search, Painter image gen, subagents
✅ Broad IDE coverage including VS Code, Cursor, JetBrains, Neovim, and Zed
✅ Remote agent execution from web and mobile with passkey auth is genuinely useful for long jobs
✅ MCP support and AGENTS.md files make it extensible and project-aware
✅ Enterprise tier offers SSO/SAML, zero data retention, and per-user cost controls

Cons

⚠️ Pay-as-you-go with no flat cap means costs can spike on large-context or agentic runs
⚠️ Frontier-model focus is expensive compared to cheaper coding assistants
⚠️ No indication of an open-source client or public API for programmatic integration
⚠️ Steeper learning curve than simple autocomplete tools; assumes agent-literate users
⚠️ $1,000 minimum for Enterprise tier is a high floor for smaller teams wanting SSO

Use cases

Multi-file refactors across large repositoriesAgentic code generation with subagent parallelismAutomated code review with custom rulesRemote background coding tasks monitored from mobileDeep reasoning on architectural or debugging problemsCross-repo code search via LibrarianCustom tool integration via MCPiOS and cross-platform app scaffoldingData-flow diagram and documentation generation