Amp

Amp is the frontier agentic coding agent built by Sourcegraph, designed to run autonomously in the terminal and editor rather than as an autocomplete assistant. Built by the Sourcegraph team led by CEO Quinn Slack, Amp deliberately optimizes for best outcomes with unfettered access to tokens and tools rather than minimizing cost, positioning it against Claude Code, OpenAI Codex, and Cursor. Amp runs multi-step coding threads, spawns subagents, and can execute agents remotely, targeting professional developers who want maximum agent capability with minimal workflow friction. Amp meets developers where they work. It ships as a CLI (installed with a single command or via Homebrew) and as extensions for VS Code, Cursor, and Windsurf, so there is no new UI to learn. Amp is multi-model, routing across frontier models including GPT-5.6 and Anthropic Claude, and exposes four effort modes (low, medium, high, and ultra) to dial how hard the agent works. It acts as an MCP client, adding tools from local or remote MCP servers, and supports a plugin system that can hook into events, add tools, standardize policy, and define custom agents and subagents. Remote execution runs agents in cloud sandboxes (orbs) with configurable CPU and memory. Amp does not expose its own MCP server, and it is not open source, so there is no self-hosted community edition. Pricing is pay-as-you-go with no subscription or commitment. Amp passes through LLM and tool costs (such as web search) with zero markup for individuals and non-enterprise workspaces, and the minimum credit purchase is $5. An Amp Free tier provides limited no-cost interactive usage. Amp Enterprise adds a 50 percent usage premium (enabled with a one-time $1,000 purchase) in exchange for SSO and directory sync, zero data retention for LLM text inputs, MCP registry allowlists, managed user settings, audit logging, and a workspace analytics API. Credits are pooled across a workspace and expire after one year of inactivity, with invoicing through Stripe. Amp is not the right fit for teams that need a fixed per-seat subscription for predictable budgeting, since usage-based pricing scales with token consumption and Amp intentionally spends tokens freely for better results. Developers who want a free, open-source agent they can self-host should look elsewhere, as Amp is proprietary. Teams wanting a lower-cost entry point or a more IDE-native experience may prefer Cursor (from $20/mo), Claude Code (from $17/mo), or OpenAI Codex (from $20/mo). Amp is also strongest for developers comfortable working in the terminal. Current state Q3 2026: Amp ships updates weekly, recently adding four tunable effort modes (low, medium, high, and ultra), remote agents runnable from anywhere the amp command runs, larger orb sizes, and plugin-created custom agents. Amp holds a 4.5 out of 5 rating across 91 G2 reviews and is SOC 2 Type II certified. Neither Amp nor its subprocessors train on customer data unless a user explicitly opts in, and training can never be enabled on Enterprise workspaces. Amp is available in the CLI, VS Code, Cursor, and Windsurf.

Free · Every Two Weeks

AI Agent Price & Rating Tracker

Price changes, new agent launches, acquisitions, and rating updates across 330+ AI agents, verified against live vendor data every 14 days.

No spam. Unsubscribe anytime. We never share your email.