Sourcegraph · Tools

Amp

Visit website

At a glance

Price
Pricing on request
Vendor
Sourcegraph

Specifications & properties

Key decision factors

Pricing model
Freemium 1
Tool form
  • CLI agent
  • IDE plugin
  • Web app
1
Autonomy level
Autonomous agent 1
Model choice
Multiple models 1

Pricing

Free tier
Yes 1

Model

Available models
GPT-5.6, Claude Fable 5 and other fast models 1

Capabilities

Repo-wide context
Yes 1
Terminal/command execution
Yes 1

Integration

IDE integrations
  • VS Code
  • JetBrains
  • Neovim
1
MCP support
Yes 1
Report data / suggest a correction

Agentic coding tool from Sourcegraph that runs on the web, in the terminal and in editors, with usage-based pricing.

Profile

Amp is an agentic coding tool that lets developers run AI coding agents directly from the terminal, VS Code, JetBrains, Neovim, Zed, or the web, with agents able to keep working unsupervised on remote machines even after a laptop is closed. It was created inside Sourcegraph and, since December 2, 2025, has been developed by the independent company Amp, spun out as "Amp Frontier Corporation" and co-founded by 20 former Sourcegraph team members including CEO Quinn Slack and Beyang Liu.

Who builds it

  • Originally built as a Sourcegraph product; Sourcegraph is the code-search and code-intelligence company founded in 2013.
  • On December 2, 2025, Amp and Sourcegraph split into two independent companies; Amp said its own traction drove the spin-out: "Amp's traction spun us out of Sourcegraph. Amp is profitable."
  • The new company was co-founded by 20 people from the Amp team, led by Quinn Slack (CEO) and Beyang Liu, both Sourcegraph co-founders.
  • Sourcegraph and Amp remain linked through shared board investors (including Craft, Redpoint, Sequoia and a16z), but operate as separate businesses with separate leadership.

Core features

  • Works as a CLI, a VS Code/VS Code-fork extension (Cursor, Windsurf), a JetBrains plugin (2025.1+), a Neovim plugin, in Zed, and on the web.
  • "Orbs": remote agent environments that keep executing tasks even when your local machine is offline, checkable from your phone.
  • Four selectable intensity modes — low, medium, high and ultra — to trade off speed against depth of reasoning.
  • An "oracle" feature that requests a second opinion from another frontier model when the primary model is stuck on a hard problem.
  • A plugin system for hooking into agent events, adding custom tools, and enforcing team policies.
  • Thread sharing and leaderboards so teams can reuse successful agent runs and monitor adoption.

Pricing

  • Amp Free — an ad-supported tier with credits that refresh hourly, worth roughly $10/day (about $300/month); ads can be switched off, in which case usage is billed at standard rates. Amp calls this "an experiment" it can't promise to run forever.
  • Pay-as-you-go — no subscription; a $5 minimum credit purchase; Amp charges "zero markup" on the underlying model providers' API pricing for individuals and non-enterprise teams; unused credits expire after a year of inactivity and are pooled across a workspace.
  • Enterprise — usage is priced roughly 50% above individual/team rates, with a $1,000 minimum purchase; enterprise customers get zero data retention for LLM text inputs and can enforce MCP-server allowlists and other policies centrally.

Who it's for

Amp suits developers who want a model-agnostic, unopinionated agent that plugs into an editor or terminal they already use rather than a brand-new IDE, and teams that want shared, auditable agent threads instead of ad hoc individual usage. Because pricing tracks raw model cost rather than a flat subscription, it also appeals to users who want to avoid paying for unused capacity.

Frequently asked questions

Is Amp free to use?

Yes — Amp Free is an ad-supported tier that gives every user roughly $10 of credits per day (about $300/month), refreshed hourly, with no payment required. Ads are text-only and Amp says they never influence the agent's responses; users who prefer no ads can turn them off and pay standard inference rates instead. Amp describes Free as an ongoing experiment rather than a permanent guarantee.

What does Amp cost beyond the free tier?

Amp is pay-as-you-go with no subscription: individuals and non-enterprise teams pay the underlying model providers' API price with zero markup, starting from a $5 minimum credit purchase. Enterprise usage costs roughly 50% more than individual/team rates and requires a $1,000 minimum purchase. Unused credits expire after a year of inactivity, and workspace credits are pooled among members.

What editors and platforms does Amp support?

Amp runs as a CLI, a VS Code (and VS Code-fork) extension, a JetBrains plugin for versions 2025.1 and above, a Neovim plugin, inside Zed, and through a web interface, plus a mobile view for checking on running agents. It supports macOS, Linux and Windows (via WSL, with WezTerm or Alacritty recommended over Windows Terminal for full clipboard support).

Is Amp still a Sourcegraph product?

No — Amp was created inside Sourcegraph but became an independent company on December 2, 2025, when Amp and Sourcegraph split into two separate businesses. The new company, founded by 20 members of the former Amp team including CEO Quinn Slack and Beyang Liu, said "Amp's traction spun us out of Sourcegraph. Amp is profitable." Sourcegraph and Amp still share some board investors but now operate independently.

Does Amp support autonomous, long-running agents?

Yes — Amp agents can run in "orbs," remote machines that keep executing a task even after you close your laptop, and you can start a task on one device (web, terminal or phone) and pick it up on another. A built-in "oracle" feature can also call a second frontier model in for a second opinion on hard problems. Plugins let teams add custom tools and hook into agent events for extra automation.

How does Amp handle data privacy for enterprise customers?

Enterprise customers get zero data retention for text inputs sent to LLM inference, according to Amp's own manual. Amp does not ask for approval before running tools by default, though plugins and configuration can change that, and users can disable individual tools or MCP servers. Enterprise admins can also deploy system-level managed policies, including MCP-server allowlists.