Testing Augment Code: A Powerful New Rival to Cursor
The article evaluates Augment Code, an AI‑powered coding assistant with 200K token context, persistent memory, multimodal input, and top SWE‑bench scores, walks through its installation, explores its use on vllm and PagedAttention, demonstrates adding a new model and auto‑generating a WeChat mini‑program, and compares its capabilities and speed to Cursor.
Augment Agent Overview
Augment Agent is an AI coding assistant designed for large codebases, supporting a context window of up to 200 K tokens. It maintains persistent memory that adapts to a developer’s coding style and conventions across sessions. Multimodal inputs such as screenshots and Figma files are accepted for error fixing and UI implementation. On the SWE‑bench verified benchmark, the combination of Anthropic Claude Sonnet 3.7 and OpenAI O1 models achieved the top score of 65.4 % on real‑world tasks.
Installation
Three usage modes are provided: a vscode extension, a JetBrains IDE plugin, and a VIM plugin. The vscode extension installs successfully; remote login fails while local login works.
Use Case 1 – Exploring vllm
After importing the vllm project, Augment Agent indexes the entire repository and explains the core of PagedAttention:
Splits loading of Attention from a contiguous GPU memory region into page‑wise loads, improving GPU memory utilization.
Implementation resides in vllm/attention/backends/abstract.py with multiple concrete subclasses (e.g., CPUMLABackend).
Two execution modes are exposed: Decode (function forward_decode) and Prefix (function forward_prefix).
Broad hardware support is also mentioned.
Use Case 2 – Adding a New Model to vllm
The task was to integrate the Salesforce SFR‑Embedding‑Code‑2B_R model. In Agent mode the assistant:
Downloaded config.json from the Hugging Face URL, parsed the architectures field, and described the model architecture.
Generated an execution plan with four steps, including implementation code and a corresponding test file.
When the generated code was executed, runtime errors occurred, attributed to mismatches with the current vllm version and incorrect import paths.
Use Case 3 – Generating a WeChat Mini‑Program
Given a specification for a family book‑management system, the assistant produced:
Front‑end code for three pages (home, book entry, personal).
A Go backend service exposing the required APIs.
A MySQL schema; the assistant launched a Docker container to create the tables.
The generated project built and ran; the front‑end operated correctly while the backend API call failed.
Comparison with Cursor
Augment Agent performed a broader set of tasks than Cursor and completed them noticeably faster, though the exact cause (model speed or other factors) was not determined.
References
[1] abstract.py: http://abstract.py
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
Infra Learning Club
Infra Learning Club shares study notes, cutting-edge technology, and career discussions.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.