Amp
Amp — the AI coding agent built for serious work
Amp pairs frontier reasoning models with the tools real engineering teams rely on. Write, refactor, and ship code with an agent that thinks across whole codebases and keeps moving when other tools stall.
~/project $ amp ▸ Analysing 248 files across 12 packages… ▸ Drafting refactor for auth/session.ts ▸ Running tests · 42 passed › Ready. What should we build next?
An agent built for engineers, not demos
Four principles shape how Amp behaves on every task — from a one-line fix to a multi-day refactor.
Frontier models
Amp uses the best models available, period. No bait-and-switch to weaker tiers when load spikes — you get frontier reasoning on every prompt.
- Top-tier reasoning by default
- Routed per task, not per plan
- No silent downgrades
Only what we love
Every feature is opinionated. The team ships what they use themselves and removes anything that doesn’t earn its place in a real engineering workflow.
- No bloated settings panels
- Sharp defaults, fewer choices
- Built by engineers, for engineers
Fast on long threads
Amp stays responsive deep into multi-step sessions. Long context, parallel tool use, and smart caching keep latency low even when the work gets ambitious.
- Streaming, parallel tool calls
- Persistent context across turns
- No degradation on long runs
Drop Amp into any workflow
Install in seconds. Use it in your editor, your terminal, or as part of CI — the agent is the same everywhere.
# Install Amp $ curl -fsSL https://ampcode.com/install.sh | bash # Authenticate and start $ amp login $ amp ▸ Amp ready. Type your task or 'help'.
Why teams choose Amp
Built for organisations that need more than autocomplete — and refuse to settle for tools that fall over on real codebases.
Works on real codebases
Handles large, multi-package repos without losing the thread — designed for production code, not toy examples.
Editor & terminal native
Use Amp inside your IDE or straight from the command line. Same agent, same behaviour, wherever you work.
Extensible toolchain
Bring your own tools, scripts, and MCP servers. Amp slots into the stack you already have.
Transparent reasoning
See what the agent is planning before it acts. Approve, edit, or redirect at any step.
Pay as you go
Usage-based pricing with no markup for individuals — no surprise overages, no opaque “credits”.
Built by an independent lab
Amp is an independent agent research lab that ships product, not papers — iterating fast at the model↔product frontier.
Where Amp pays off
Amp is built for the kinds of work where shallow autocomplete runs out of road.
Large refactors
Plan and execute multi-file changes that touch dozens of modules without losing context.
Bug investigation
Trace failures across the stack — Amp reads logs, runs tests, and proposes targeted fixes.
Greenfield builds
Spin up new services and features with sensible scaffolding and tests from the first commit.
Codebase onboarding
Get up to speed on unfamiliar code — ask questions, walk the call graph, and document as you go.