Amp Code Free: The AI Coding Agent That Works in Your Editor
Discover Amp Code, the powerful AI coding agent from Sourcegraph. Learn about Amp Free with unlimited access, the CLI tool, IDE integration, and how it compares to paid alternatives.
Table of Contents
- What is Amp Code?
- Amp Free vs Smart Mode
- Amp IDE Extension: VS Code, Cursor, and Windsurf
- Amp CLI: Terminal-Based AI Assistant
- Advanced Features
- Real-World Example Prompts
- Workspaces: Team Collaboration
- Comparison with Other AI Coding Tools
- Getting Started with Amp
- Integrations and Extensions
- Frequently Asked Questions
- Conclusion
Looking for a powerful AI coding agent that actually understands your codebase? Amp Code by Sourcegraph is a frontier coding agent that runs directly in your editor and terminal, giving you unconstrained access to the best AI models. Even better, Amp Free mode offers unlimited access supported by ads, making advanced AI coding assistance available to everyone.
What You'll Discover
- Free unlimited AI coding with Amp Free mode (ad-supported)
- VS Code, Cursor, and Windsurf integration for seamless workflow
- Powerful CLI tool for terminal-based development
- Advanced features like Oracle (GPT-5), subagents, and custom tools
- Smart mode with unconstrained token usage for complex projects
If you’re exploring AI tools for development, you might also want to check out how to use Claude Sonnet 4.5 and GPT-5 for free or learn about building AI affiliate websites.
What is Amp Code?
Amp is a frontier coding agent built by Sourcegraph that brings AI assistance directly into your development workflow. Unlike simple code completion tools, Amp is designed as a true agent that can understand your entire codebase, execute commands, edit files, and work autonomously on complex tasks.
Core Principles
Amp is built around four key principles that make it different from other AI coding tools:
- Unconstrained token usage: No artificial limits on context or output. If your task needs millions of tokens, Amp uses them.
- Always the best models: You don’t pick models—Amp automatically uses the most capable models available (currently Claude Sonnet 4.5).
- Raw model power: Direct access to what frontier models can actually do, not watered-down versions.
- Built to evolve: Automatically stays current with the best models as they’re released.
Amp Free vs Smart Mode
Amp offers two distinct modes to fit different needs and budgets:
Amp Free Mode
Amp Free is completely free of charge, supported by ads and training data sharing. It uses a mix of top open-source models, frontier models with limited context windows, and pre-release models in testing.
Key Features:
- Completely free to use for work or personal projects
- Unlimited usage (some rate limits apply)
- Ad-supported with tasteful developer-focused advertisements
- Training data sharing required (models train on your usage)
- Interactive use in editor or CLI
- Not available for enterprise workspaces
Training Data Requirement
To make Amp Free sustainable, your code and conversations are used to train models. If you’re working with proprietary code or don’t want this, use Amp Smart mode instead.
Amp Smart Mode (Paid)
Smart mode is Amp’s default paid tier that offers the full frontier model experience with unconstrained usage.
Key Features:
- Claude Sonnet 4.5 with up to 1 million token context
- Unconstrained token usage for complex tasks
- No training data sharing (zero data retention available for Enterprise)
- Oracle access (GPT-5 for complex reasoning)
- Subagents for parallel task execution
- Custom tools via Model Context Protocol (MCP)
Pricing:
- Pay-as-you-go based on actual model usage
- Direct pass-through of API costs with no markup
- Most users get $10 USD in free credits to start
- Workspace credits pool usage across team members
Mode | Cost | Models | Training | Best For |
---|---|---|---|---|
Free | $0 | Mixed OSS + Frontier | Required | Learning, personal projects |
Smart | Pay-as-you-go | Claude 4.5, GPT-5 | Optional | Professional development |
Enterprise | 50% markup | Best available | Never | Teams, proprietary code |
Amp IDE Extension: VS Code, Cursor, and Windsurf
The Amp IDE extension integrates directly into your code editor, providing a seamless AI assistant experience. It works with VS Code and compatible forks like Cursor and Windsurf.
Installation
Installing Amp in your IDE is straightforward:
For VS Code:
- Visit ampcode.com and sign in
- Install the Amp extension from VS Code Marketplace
- Authenticate with your Amp account
- Start coding with AI assistance
For Cursor or Windsurf: The same extension works in these VS Code-compatible editors. Follow the installation instructions on the Amp dashboard.
Key Features in the IDE
Amp organizes conversations as threads—think of them like Git branches for AI conversations. Each thread maintains its own context, file changes, and conversation history.
- Create new threads for different tasks or features
- Switch between threads to work on multiple things
- Share threads with your team (if in a workspace)
- Track file changes made by the agent
- Revert changes individually or all at once
Amp automatically looks for AGENTS.md
files in your project to understand:
- Build and test commands
- Architecture and conventions
- Common pitfalls to avoid
- How to run and review code
Place AGENTS.md
in your project root or subdirectories. Amp includes them automatically when working in those areas. You can also use $HOME/.config/AGENTS.md
for personal preferences.
Example AGENTS.md
:
# Project Guide for AI Agents
## Commands
- **dev**: `npm run dev`
- **build**: `npm run build`
- **test**: `npm test`
## Architecture
- React + TypeScript frontend
- Node.js + Express backend
- PostgreSQL database
## Important Notes
- Always run tests before committing
- Use TypeScript strict mode
- Follow existing code patterns
Amp tracks your context window usage and provides tools to manage it:
- Hover over context indicator to see which files are loaded
- Compact Thread to summarize and reduce context usage
- New Thread with Summary for fresh start with context carryover
- File Changes view shows what the agent modified
- View
AGENTS.md
files currently in use with/agent-files
Practical Example: Building with Amp
Here’s how a typical Amp workflow looks in your IDE:
You: "Fix all the TypeScript errors in this file"
Amp:
- Reads the file and diagnostics
- Identifies the type errors
- Fixes each error with proper types
- Runs TypeScript compiler to verify
- Shows you the changes made
Or for more complex tasks:
You: "Look at localhost:3000 and make the header more minimal"
Amp:
- Takes a screenshot of the localhost URL
- Analyzes the current header design
- Suggests improvements
- Modifies the CSS/components
- Takes another screenshot to verify
- Iterates until it looks right
Pro Tip
Use Cmd/Ctrl+Enter
to submit messages in Amp. This deliberate action encourages you to write better prompts and get better results.
Amp CLI: Terminal-Based AI Assistant
The Amp CLI brings AI coding assistance directly to your terminal—perfect for developers who live in the command line or need to integrate AI into scripts and workflows.
Installation
Installing the CLI is simple:
# Using the install script (Linux, macOS, WSL)
curl -fsSL https://ampcode.com/install.sh | bash
# Or using npm
npm install -g @sourcegraph/amp
# Or using pnpm
pnpm add -g @sourcegraph/amp
# Or using Yarn
yarn global add @sourcegraph/amp
Interactive Mode
Run amp
without arguments for interactive mode:
$ amp
# Will prompt for login on first run
# Then start an interactive session
You can pipe input to start with context:
$ echo "commit all my changes" | amp
Execute Mode
Use -x
or --execute
for non-interactive mode (requires paid credits):
$ amp -x "what files in this folder are markdown files?"
README.md
AGENTS.md
docs/getting-started.md
Combine with pipes for powerful workflows:
$ cat package.json | amp -x "what package manager is used?"
npm
$ git diff | amp -x "write a commit message for these changes"
fix: resolve TypeScript errors in authentication module
Slash Commands
The CLI supports powerful slash commands:
/help
- Show help and hotkeys/new
- Start a new thread/continue
- Continue an existing thread/quit
- Exit Amp
/generate-agent-file
- Create AGENTS.md for your project/agent-files
- List AGENTS.md files in use/permissions
- Edit permission rules/compact
- Reduce context usage
/queue [message]
- Queue message for later/dequeue
- Restore queued messages/editor
- Open $EDITOR to write prompt
Custom Slash Commands
Create your own slash commands by adding scripts to .agents/commands
or ~/.config/amp/commands
:
Markdown files become prompt templates:
# .agents/commands/pr-review.md becomes /pr-review
Executables run and send output to Amp:
#!/usr/bin/env bash
# ~/.config/amp/commands/outline
# Becomes /outline command
tree -L 2 "$@"
Shell Mode
Execute shell commands directly in the CLI:
# Regular shell mode (included in context)
.ls -la
# Incognito mode (not included in context)
$pwd
IDE Integration
The CLI can connect directly to your IDE:
# For VS Code or Neovim with Amp installed
$ amp --ide
# For JetBrains IDEs
$ amp --jetbrains
This gives Amp access to:
- Currently open files
- Selected code
- IDE diagnostics
- Direct file editing with undo support
Advanced Features
Oracle: Second Opinion from GPT-5
Amp includes access to a powerful “Oracle” tool that uses GPT-5 for complex reasoning and analysis. While Claude Sonnet 4.5 is great for day-to-day coding, sometimes you need deeper analysis.
Example usage:
"Use the oracle to review the last commit's changes. I want to make sure
the actual logic hasn't changed, only the implementation."
"Analyze how these two functions work, then ask the oracle to figure out
how we can refactor the duplication while keeping it backwards compatible."
The main agent decides autonomously when to consult the oracle, but you can explicitly request it for better results on complex problems.
Subagents for Parallel Work
Amp can spawn subagents to work on independent tasks in parallel. Each subagent has its own context window and can use tools like file editing and terminal commands.
When to use subagents:
- Converting multiple files to a new pattern (e.g., CSS to Tailwind)
- Running tests and fixing failures across different modules
- Independent feature implementations
- Tasks with extensive output you don’t need in main context
Example:
"Use 3 subagents to convert these CSS files to Tailwind. One subagent per file."
Each subagent works independently and reports back when done, keeping your main thread clean and focused.
Custom Tools with MCP
Amp supports the Model Context Protocol (MCP) for adding custom tools. Configure MCP servers in your settings:
"amp.mcpServers": {
"playwright": {
"command": "npx",
"args": ["-y", "@playwright/mcp@latest", "--headless"]
},
"linear": {
"url": "https://mcp.linear.app/sse"
}
}
This lets Amp interact with external services, APIs, and tools specific to your workflow.
Amp Tab: AI-Powered Completions
Amp Tab is an experimental completion engine that anticipates your next actions:
- Multi-line code suggestions
- Edits in other files based on context
- Semantic understanding of your changes
- Language server diagnostics integration
Enable in settings: "amp.tab.enabled": true
Press Tab
to accept suggestions and jump to additional edits.
Real-World Example Prompts
Here are concrete prompts you can try with Amp:
"Fix all the TypeScript errors in this file"
"Run the tests and fix any failing ones"
"Review this API design and suggest improvements" (uses Oracle)
"Add a dark mode toggle to this React component"
"Plan how to add real-time chat to this app, but don't write code yet"
"Find where user authentication is handled in this codebase"
"git blame this file and tell me who added that function"
"Check git diff --staged and remove the debug statements"
"Look at the last commit and help me change this feature"
# Command line usage
amp -x 'what files in this folder are markdown?'
# With context from commands
git log --oneline -10 | amp -x "summarize recent changes"
# Visual debugging
amp -x "Look at localhost:3000 and make the header more minimal"
Workspaces: Team Collaboration
Amp Workspaces provide collaborative environments where teams can share knowledge and threads.
Key Features
- Thread sharing - All threads visible to workspace members by default
- Pooled billing - Shared credit pool for the team
- Leaderboard - Track activity and contributions
- Learning from others - Browse and learn from teammate’s threads
- SSO support - Enterprise workspaces can enable single sign-on
Privacy Controls
Threads can be:
- Workspace-shared (default for workspace members)
- Private (only visible to you)
- Public (visible to anyone with link)
Change visibility anytime through the sharing menu.
Comparison with Other AI Coding Tools
Feature | Amp Free | Amp Smart | Other Free Tools |
---|---|---|---|
Cost | Free (ads) | Pay-as-you-go | Free with limits |
Context Window | Limited | 1M+ tokens | Usually limited |
Model Selection | Auto (mixed) | Claude 4.5 | User picks |
Training Data | Required | Optional | Varies |
IDE Integration | ✓ | ✓ | Limited |
CLI Tool | ✓ | ✓ | Rare |
Subagents | ✗ | ✓ | ✗ |
Oracle (GPT-5) | ✗ | ✓ | ✗ |
Custom Tools | Limited | ✓ | ✗ |
Enterprise | ✗ | ✓ | ✗ |
Getting Started with Amp
Ready to try Amp? Here’s your roadmap:
Step 1: Choose Your Mode
For Learning & Personal Projects:
- Start with Amp Free mode
- Get unlimited usage with ads
- Perfect for experimenting and learning
For Professional Development:
- Use Amp Smart mode
- Get $10 free credits to start
- Pay only for what you use
Step 2: Install Amp
IDE Extension:
- Visit ampcode.com and sign up
- Install extension for VS Code, Cursor, or Windsurf
- Authenticate and start coding
CLI Tool:
curl -fsSL https://ampcode.com/install.sh | bash
amp # Login and start
Step 3: Create AGENTS.md
Help Amp understand your project:
# In your project root
amp /generate-agent-file
Or create manually with:
- Build/test commands
- Architecture overview
- Important conventions
- Common pitfalls
Step 4: Start Coding
Try these starter prompts:
"Run the tests and fix any failures"
"Look at this file and explain what it does"
"Help me add error handling to this function"
Best Practices
- Be explicit - Instead of “can you do X?”, say “do X”
- Keep it focused - One task per thread works best
- Provide context - Tell Amp which files or commands matter
- Use AGENTS.md - Guide Amp on how to test and build
- Start fresh - New thread if context gets cluttered
- Review work - Tell Amp how to verify its changes
Integrations and Extensions
Amp integrates with many other AI and development tools you might already be using:
- Learn about MCP servers in BrightData for web scraping capabilities
- Explore best open-source LLMs as Claude alternatives
- Check out other AI tools and resources for development
Frequently Asked Questions
Yes, Amp Free offers unlimited interactive usage with some rate limits to prevent abuse. You can use it as much as you need for personal projects and learning. The free tier is supported by ads and training data sharing.
In Amp Free mode, your code and conversations are used to train AI models. This is how the free tier remains sustainable. If you’re working with proprietary or sensitive code, use Amp Smart mode instead, which offers zero data retention options.
Amp Smart uses pay-as-you-go pricing based on actual model usage. Costs are passed through directly with no markup for individuals and teams (50% markup for Enterprise). Most complex tasks cost a few cents to a few dollars depending on token usage.
Yes! Switch modes anytime with /mode free
or /mode smart
in the CLI, or select the mode in your IDE’s prompt field. Your threads and history are preserved across modes.
No, Amp requires an internet connection to access AI models. However, the IDE integration and file editing work locally, so you maintain full control of your code.
Amp works with VS Code, Cursor, Windsurf, and other VS Code-compatible editors. There’s also experimental support for Neovim and JetBrains IDEs (IntelliJ, WebStorm, GoLand, etc.).
Conclusion
Amp Code brings frontier AI capabilities directly into your development workflow—whether you prefer working in an IDE or the command line. With Amp Free, you get unlimited access to powerful AI coding assistance without spending a cent, while Amp Smart offers unconstrained access to the best models for professional development.
Key Takeaways:
✅ Amp Free - Unlimited AI coding with ads, perfect for learning ✅ Amp Smart - Unconstrained Claude 4.5 access, pay only what you use ✅ IDE Integration - Works in VS Code, Cursor, Windsurf seamlessly ✅ Powerful CLI - Terminal-based AI for scripts and automation ✅ Advanced Features - Oracle (GPT-5), subagents, custom tools via MCP ✅ Team Collaboration - Workspaces with thread sharing and pooled billing
Whether you’re building side projects, learning to code, or working on production applications, Amp provides the AI assistance you need at a price point that works for you.
Get Started with Amp CodeContinue Learning
Explore more AI tools and development resources:
Ready to experience the future of AI-powered coding? Install Amp today and start building with the assistance of frontier AI models!
Related Posts

Best AI Coding Tools and Assistants in 2025
A comprehensive guide to the most effective AI coding tools and assistants in 2025, including web tools, IDE integrations, and CLI solutions.

Podman vs Docker - Which Container Tool Should You Choose in 2025-2026?
Complete comparison of Podman and Docker container engines. Learn about security, performance, and which tool fits your development needs best.
Best Open Source LLMs to Replace Claude Sonnet 4.5: Affordable AI Coding Alternatives 2025
Discover the top 3 open source language models that can replace Claude Sonnet 4.5 for coding tasks at a fraction of the cost: GLM-4.6, Kimi K2-0905, and Qwen-Max.