Augment Agent is the AI pair programmer that deeply understands your codebase. Uses Memory, Tools (MCP & Native), 200K context. #1 SWE-bench approach open-sourced. Available for VS Code and JetBrains.
Hi everyone!
Check out Augment Agent, a new AI pair programmer designed specifically for the challenges of real-world software engineering, especially within large, complex codebases where many AI tools often fall short.
Augment's core focus is deep context. Their "Context Engine" aims to give the agent a comprehensive understanding of your entire engineering ecosystem – code history, documentation, tickets, even past conversations. It also introduces:
🧠 Memories: The agent learns your coding style, preferences, and project specifics over time, persisting across sessions.
🛠️ Extensive Tool Integration: Supports the Model Context Protocol (MCP) for broad connectivity and includes built-in Native Tools for essential platforms like GitHub, Jira, Notion, and Linear.
↔️ Large Context Window: Handles up to 200K tokens, crucial for complex tasks.
📸 Multimodal Input: Accepts screenshots and Figma files to help with UI implementation or bug fixing.
⏪ Code Checkpoints: Automatically saves progress for easy rollbacks if needed.
🤖 Auto Mode: An option for the agent to operate more autonomously.
Impressively, the team achieved the #1 spot on the SWE-bench verified leaderboard using a specific agent setup (combining Claude 3.7 and O1) and has open-sourced their implementation for that benchmark, showing how they did it.
Augment Agent integrates with VS Code and JetBrains.
