Bug hunting with Amazon Bedrock and SWE-Agent 👨‍💻

Amazon Bedrock ❤️ SWE-Agent

I'm very pleased to announce that starting with the 0.3.0 release you can now run SWE-Agent using Amazon Bedrock ⛰️🎉

❗ As of May 2024, Amazon Bedrock support in SWE-Agent only covers Anthropic Claude models.

What is SWE-Agent? 👨‍💻

SWE-Agent is an open source project developed and maintained by the Princeton NLP group that turns language models into software engineering agents that can fix real-life bugs and issues in GitHub repositories.

Based on a Agent-Computer Interface (ACI) paradigm, SWE-Agent is able to resolve ~12.29% of issues in the SWE-Bench dataset and takes just a few minutes to run.

Getting Started 🚀

0a/ Enable access to Anthropic’s Claude models via Amazon Bedrock

💡 Please refer to the Amazon Bedrock User Guide (Set up > Model access) for more information on how to request model access.

0b/ Install Docker 🐳

1/ Set up AWS credentials

🔒 For security reasons, SWE-Agent will not look inside the keys.cfg file for AWS credentials.

1
2
3
export AWS_ACCESS_KEY_ID=...
export AWS_SECRET_ACCESS_KEY=...
export AWS_SESSION_TOKEN=...

2/ Pull the SWE-Agent image

1
docker pull sweagent/swe-agent:latest

3/ Run SWE-Agent

☝️ Use the --temperature and --top_p options to have more control over the model output.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
# ❗ As of May 2024, Bedrock support only covers Claude models
MODEL_ID=anthropic.claude-3-sonnet-20240229-v1:0

# What we want to fix
GITHUB_ISSUE=https://github.com/pvlib/pvlib-python/issues/1603

# Start SWE-Agent
docker run --rm -it -v /var/run/docker.sock:/var/run/docker.sock \
					$(env | grep AWS_ | xargs printf ' -e %s') \
					sweagent/swe-agent-run:latest \
					python run.py --image_name=sweagent/swe-agent:latest \
								  --model_name bedrock:$MODEL_ID \
								  --data_path $GITHUB_ISSUE \
								  --config_file config/default_from_url.yaml \
								  --skip_existing=False

Among other things, SWE-Agent will save the thought-action-observation steps generated by the agent when trying to solve the issue inside a JSON-formatted trajectory file, which you can visualize with the inspector tool:

Depending on the model selected and the inference parameters, SWE-Agent can even create a patch file that you can apply directly to a local repository:

Try it out and let me know what you think in the comments section below 👇

Any opinions in this post are those of the individual author and may not reflect the opinions of AWS.

Select your cookie preferences

Site Terms, Privacy, and more.

Bug hunting with Amazon Bedrock and SWE-Agent 👨‍💻

Use Amazon Bedrock with SWE-Agent to create your own software engineering agent that can fix real-life bugs and issues in GitHub repositories.

Amazon Bedrock ❤️ SWE-Agent

What is SWE-Agent? 👨‍💻

Getting Started 🚀

Comments