Awesome/awesome-ai-agents
1
0
Fork 0
mirror of https://github.com/e2b-dev/awesome-ai-agents.git synced 2026-02-22 19:35:12 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add SWE-agent

Browse Source
This commit is contained in:
Tereza Tizkova
2024-04-10 16:04:23 -07:00
parent 75af3265e5
commit 69fae09395
1 changed files with 16 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
README.md
View File

@@ -2356,12 +2356,13 @@ General purpose
</details>
## [SWE Agent]()
Open Source alternative for Devin
## [SWE Agent](https://github.com/princeton-nlp/SWE-agent)
Open-source Devin alternative
<details>
![Image](https://github.com/stitionai/devika/raw/main/.assets/devika-screenshot.png)
![Image](https://www.swebench.com/img/swellama.png)
### Category
Coding, general purpose
@@ -2374,11 +2375,21 @@ Coding, general purpose
- 🧪 Write and execute tests
- This custom-built interface is critical for good performance. Simply connecting an LM to a vanilla bash terminal does not work well.
- ["Our key insight is that LMs require carefully designed agent-computer interfaces (similar to how humans like good UI design). E.g. When the LM messes up indentation, our editor prevents it and gives feedback."](https://x.com/jyangballin/status/1775114448513958134)
- SWE-agent was released by the Princeton NLP team.
- What makes SWE-agent special is that it performs almost as well as Devin on the SWE-bench.
- It is important to say that the performance [varies](https://www.swebench.com/) based on the model used by the agent.
- The changes and innovations in SWE-agent compared to Devin are:
- The code in SWE Agent is executed locally via Docker.
- It uses “Agent-Computer Interface” (ACI) - constraining the interface makes the agent easier to use for LMs. Only a few commants are allowed: run code, look for code, edit code and submit changes to GitHub.
- Any code the agent writes goes through a syntax check (linter) before being submitted. If the syntax is incorrect, the agent gets feedback and is forced to redo the code.
- The agent can only read 100 lines of code at a time, rather than the entire file. This makes it easier for the language model to understand the code.
### Links
- [GitHub]()
- [X (Twitter) post](https://x.com/Gautam_A_k/status/1775206295458676764)
- [GitHub](https://github.com/princeton-nlp/SWE-agent)
- [Web](https://swe-agent.com/)
- [Demo](https://swe-agent.com/demo)
- [Discord](https://discord.com/invite/AVEFbBn2rH)
</details>