Fume

AI Agent for Automated Code Testing and Review

Free Trial

Description

Fume introduces an AI-powered approach to software development, functioning as an automated system for testing and reviewing code changes before they are merged. It aims to help engineering teams increase their development speed without sacrificing code quality by identifying potential issues early in the process. The tool analyzes pull requests, understands the context of the changes, and meticulously tests for edge cases that might be overlooked during manual review.

Beyond static analysis, Fume performs dynamic testing using its own browser and command-line interface, simulating real-world interactions and providing screen recordings attached to reviews for verification. It navigates the entire codebase contextually, similar to how a human engineer would. Fume operates as an agentic tool, using a virtual computer environment to execute tests, edit files, and run commands. It can suggest multi-file, tested code changes that developers can implement quickly. Teams can customize Fume's review process by providing specific instructions through playbooks or web recordings, integrating team-specific knowledge into the automated workflow.

Key Features

Automated Edge Case Testing: Tests edge cases for each PR, understanding context around changes.
Scalable Code Review: Scales the team's ability to review code as the team grows.
Actionable Code Suggestions: Suggests multi-file, tested changes that can be committed with a single click.
Dynamic Testing: Actively tests code changes using browser and command line interactions.
Screen Recordings: Attaches screen recordings of dynamic tests to code reviews.
Full Codebase Context: Navigates the codebase like an engineer to understand context.
Agentic Computer Use: Utilizes a computer environment to run commands, edit files, and perform web actions for testing.
Customizable Knowledge: Can be trained with team-specific knowledge via playbooks or web recordings.
Multi-Platform Integration: Communicates via Web, Slack, Github, and potentially others (Gitlab, Bitbucket for Enterprise).