Changelog

Stay up to date with the latest improvements, feature updates, enhancements, and bug fixes.

FeatureDec 5, 2025

End-to-End Testing in Paragon

Paragon now supports end-to-end testing through natural language. Describe what you want to test in plain English, and Paragon generates and runs Playwright tests for you.

Features

  • Natural Language Test Prompts: Describe any flow, component, or interaction you want to test. Paragon interprets your description and generates the appropriate Playwright tests.
  • Local Environment Intelligence: Tests execute entirely on your machine, leveraging your real dev environment, local services, env vars, secrets, and custom configurations for maximum fidelity.
  • Automatic Report Generation: Paragon runs your tests and opens the Playwright HTML report so you can review results immediately.
  • Persistent Test Files: Generated test files remain in your repository, so you can inspect, modify, or extend them as needed.
CLI UpdateNov 30, 2025

Paragon v0.0.13

New features, improvements, and changes in Paragon CLI v0.0.13.

New Features

  • Slash Commands: New slash command system for faster access to common actions.
  • Image Support: Drag and drop images directly into the CLI for multimodal conversations.
  • Monitor Management: Add, edit, and delete monitors directly from the CLI.
  • Feature Mode Toggle: Switch between feature modes without leaving the terminal.

Improvements

  • Subagent UI/UX: Significantly improved interface and experience for working with subagents.
  • Better Onboarding: Clearer instructions at the start screen for new users.
  • Updated Thinking Budgets: Adjusted token budgets across tiers: max (64k), high (16k), mid (10k).
  • Fixed Context Display: Model context numbers now display correctly.

Removed

  • Old command palette (replaced by slash commands)
  • Repo indexing
  • Legacy GitHub CLI commands
Major ReleaseNov 3, 2025

Introducing Paragon - Multi-Agent QA Engineer

Paragon is a multi-agent QA system that pinpoints critical issues in your codebase directly from your terminal. Powered by parallel AI agents and deep code analysis, Paragon detects problems other tools miss. This release represents a fundamental shift in how teams approach code quality. from reactive bug fixing to proactive issue prevention.

ReviewBenchLite Accuracy Results

Paragon outperforms all competitors on the authoritative code review benchmark

81.2%
Paragon Deep
72.6%
Paragon Fast
65.8%
Greptile V3
56.4%
Claude Code
51.3%
Cursor Bugbot
44.4%
Codex
22.2%
CodeRabbit

Higher is better. Accuracy measured across 117 code review scenarios.

Features

  • Terminal-Native Code Review CLI: Detect deep-seated issues across infrastructure, security, control flow, and architecture in any part of your codebase. Comprehensive analysis without leaving your terminal.
  • Deep Research Agents: Intelligent agents that index and analyze your entire codebase to build comprehensive understanding. They reference documentation, best practices, and cross-file dependencies to uncover issues hidden in complex interactions.
  • Deep Review Mode: Spawn 8 Paragon agents in parallel to conduct exhaustive code analysis. Each agent specializes in different aspects. security, performance, architecture, testing. compiling a comprehensive, categorized issue list in minutes.
  • Automatic PR Comments: Powered by Paragon Heavy, automatically post detailed review comments on new pull requests. Issue detection happens seamlessly in your workflow. no manual intervention required.

Improvements

  • Redesigned Dashboard UI: Completely reimagined interface with streamlined workflows and intuitive navigation. Everything you need for code review at your fingertips.
  • Industry-Leading Benchmarks: Paragon outperforms all competitors on ReviewBench, the authoritative code review benchmark. Both Fast and Deep modes achieve higher accuracy than any published baseline.
  • Enhanced Detection Accuracy: Improved agent reasoning with 40% better issue identification across security vulnerabilities, performance bottlenecks, and architectural problems.
  • Faster Review Times: Optimized parallel execution reduces average review completion time by 60% compared to previous versions.