Patched

TLDR; Patched empowers developers to build and deploy AI-assisted workflows that can autonomously complete post-code tasks like documentation, reviews, and patching. Try us on our app today or use our open-source repo!

Hey everyone, we’re Asankhaya and Rohan - ex-researchers and second-time founders who want to help development teams automate tasks that stop them from shipping code faster.

Fun fact: We went to grad school at competing universities and built competing start-ups before this!

Problem

Development teams slow down not because building new features is hard but because they get overwhelmed by maintenance and patching tasks. On average, devs spend 35% of their time just managing existing code.

Solution

AI and LLMs can help, but they need to be given well-defined, tailored instructions with clear context. Patched enables developers to do this through customizable workflows called "patchflows" that combine atomic actions with AI to complete a code management task.

How does it work?

You can customize or build automations using the patchflow builder. Here’s an example of a PR Review patchflow that adds comments if there are coding style violations.

Patchflows can be triggered on your code repositories through the app, via CLI, or as part of your CI.

What makes Patched different?

• Autonomous Automation: Unlike copilots that assist as you code, Patched runs in the background, autonomously managing tasks so you don’t have to.
• Code-Native Workflows: You can build patchflows using plain text and our visual builder, but they all translate to Python code that you can inspect, edit, and extend as you like.
• Open-Source Flexibility: Our open-source framework allows you to run patchflows locally, customize them to your needs, and even integrate your own LLM APIs for enhanced privacy and control.
• Control & Precision: AI agents struggle to plan and achieve reliable outcomes, and humans struggle to provide clear instructions. Patched gives you full control over your workflows, allowing you to define, iterate, and refine until you achieve the desired outcome—no guesswork involved.

The result is a framework that can reliably automate tasks with up to 90% success rate, through efficient model use and carefully curated prompts.

Our Ask:

• ⭐ Try Patched (no waitlists!) and star our repo!
• If you’re part of a large dev team building AI tooling and workflows for yourself, we’d love to talk.