AutoCodeRover (now Sonar)

📣Big news! We’re excited to share that AutoCodeRover has been acquired by @SonarSource 🎉. Bringing our sophisticated AI agent platform into the Sonar solution, we’ll help redefine the SDLC, supercharging developers to build better, faster. More here: bit.ly/3X9UT3H

We are excited to be part of the inaugural cohort of the National GRIP Programme in Singapore (grip.sg),  organized by NRF, NUS, and NTU. #AutoCodeRover #Singapore #National #GRIP

Register for a free trial in our Cloud Platform at app.autocoderover.dev

Our latest experiments using AutoCodeRover-v2 achieves 46.2% pass@1 on SWE-Bench Verified, 37.33% on SWE-Bench Lite and 24.89% on SWE-Bench Full. Free Trial: app.autocoderover.dev GitHub: github.com/nus-apr/auto-c… #AutoCodeRover #AI4SE #AIDevAssistant

Thrilled to announce our latest development, bringing effortless automation to software development. You can seamlessly invoke AutoCodeRover directly within GitHub issues and generate software patches. Learn More: @autocoderover/f0d91ac6162c" target="_blank" rel="nofollow noopener">medium.com/@autocoderover… #AI #Autonomous #Engineering

🚀 Releasing AutoCodeRover-v2 (v20240620)! 🚀 We’re excited to release AutoCodeRover-v2, which is now more capable and open-source! Our source code is now open for non-commercial use. You can apply for access today at #application" target="_blank" rel="nofollow noopener">autocoderover.dev/#application #AutoCodeRover #OpenSource

On top of efficacy, we focus on precision - not proposing false patches to the user. AutoCodeRover-v2 is more precise than other open-source agents by filtering out incorrect patches using a reviewer agent! arXiv: arxiv.org/abs/2408.02232. And stay tuned for access in 45 days!

Beyond efficacy - AutoCodeRover-v2 is here! Augmenting AutoCodeRover with specification inference, AutoCodeRover-v2 resolves 31% of SWE-Bench lite & 19.31% of SWE-Bench, with a modest cost of ~$0.7 per task & to be open-source! Find out on arXiv: arxiv.org/abs/2408.02232 1/2

The latest version of AutoCodeRover (using GPT-4o) resolves 30.67% of the tasks (pass @1) in SWE-bench Lite. Achieving this efficacy while economical with only 0.12m tokens costing $0.7 per task and completing each task within 7 mins. #AutoCodeRover #AISE #AIDeveloper #SWEBench