All Hands AI
@allhands_ai
Followers
3K
Following
180
Media
13
Statuses
161
We build AI software development agents, in the open. Developing OpenHands (prev: OpenDevin): https://t.co/wDOBeXGLmO
Joined May 2024
We just released, open source, the strongest AI software developer to date:. 🥇 53% resolve rate on SWE-Bench Verified.🥇 41.7% resolve rate on SWE-Bench Lite. Details: 🧵↓
7
25
129
In this work with UC Berkeley and CMU, we found that o1 underperforms Claude when used in OpenHands. We're still digging into details, but we thought it'd be interesting to share our intermediate results now!. Reasoning is not a panacea for agents, other elements are necessary.
Surprising find: OpenAI's O1 - reasoning-high only hit 30% on SWE-Bench Verified - far below their 48.9% claim. Even more interesting: Claude achieves 53% in the same framework. Something's off with O1's "enhanced reasoning". 🧵1/8
2
8
62
We are proud to announce that All Hands has raised $5M to build the world’s best software development agents, and do it in the open 🙌. Thank you to @MenloVentures and our wonderful slate of investors for believing in the mission!.
All Hands AI raises $5M to build open source agents for developers
3
6
59
Last week we announced that OpenHands is the strongest software developer in the world. Today we're officially announcing an online app that makes OpenHands easier to use than ever: It's currently in beta and rolling out quickly, sign up today!.
We just released, open source, the strongest AI software developer to date:. 🥇 53% resolve rate on SWE-Bench Verified.🥇 41.7% resolve rate on SWE-Bench Lite. Details: 🧵↓
5
11
56
We're excited to end 2024 with OpenHands becoming the 70th most starred Python repo on GitHub! We’ve come so far in these 9 months or so, and thanks a bunch to the wonderful maintainers, contributors, and community that brought us here 🚀
2
8
56
LLM-powered coding agents are all the rage, but how good are various LLMs when used as coding agents?.To figure this out, we:.- Developed cloud-based infra for agent eval that speeds evals by 30x or more.- Used this to evaluate 10 SOTA open & closed models.
2
7
52
OpenHands now the 90th most starred Python repo on Github 🌠. It’s truly an honor to be listed with other great libraries like the OpenAI gym, LLaMA-Factory, ray, and gradio! Really proud of our community and everything we’ve done, let's continue to go🚀.
2
5
48
Announcing the release of OpenHands 0.16.0 🙌. OpenHands now authenticates to your repositories as a github app. This means that (1) no more need to get a github personal access token, and (2) you can control which repos OpenHands has access to!.
2
4
37
You're sleeping on AI SWEs: Nice blog post about why agents for AI development can greatly speed productivity over the standard workflow (using OpenHands as an example of course 😀).
1
3
35
Say hello to our new name, OpenHands!. If you'd like to help us build software development agents in an open and collaborative way:.* Join our open source community: * Apply for our open positions:
Announcement: the maintainers team of the OpenDevin open AI software development agent have decided to rename the project to OpenHands 🙌. We have also made a big 0.9.0 release with a number of new features, read below for details.
1
5
31
📈So far, in November, OpenHands authored or co-authored ~60% of commits to the openhands-resolver repo and ~20% of the main repo
1
2
31
📣Announcing OpenHands 0.10.0!📣. Our biggest release in a while with.- A brand new UI .- Ability to connect to Github projects w/ tokens.- Getting started examples.- Many resiliancy improvements.- Support for running sandboxes on @modal_labs .
2
4
30
OpenHands 0.21.0 is out! Some big changes 🎉. * OpenHands can now be used in 10 languages (e.g. Chinese, Japanese, Arabic, Spanish).* You can use OpenAI o1 and set reasoning_effort.* You can use a separate LM as a draft editor.
1
5
31
We're excited to announce that the first OpenHands meetup will take place in NYC 🗽 on Monday, Feb 24!. You can sign up here: Space is limited!. We're also planning to do something in SF 🌉 so stay tuned as well.
2
8
28
We can't help but agree 🫡.
2
1
25
Exciting release of OpenHands 0.14.0:. * We have ⭐️integrated VSCode!⭐️ - you can now an IDE to edit files directly from the OpenHands interface.* Better support for non-function-calling LMs.* Better support for long sessions.* Robustness improvements.
1
6
21
Big release of OpenHands 0.15.0 🙌 (so big we need two posts 🚀). 1/2.* More efficient web browsing through markdown.* A web browsing alignment checker to prevent dangerous web actions (w/ @invariant_labs).* Custom docker images for the github resolver.
2
1
20
Announcing OpenHands 0.9.5!. - Modifications to support OpenAI o1 models.- Added better error displays.- Many improvements to UI-based configuration.- Improved responsiveness to users starting/stopping the agent.
0
4
19
Want to use AI agents in your development workflow? We released a tool that lets you.- Add a github action where you tag an issue with "fix-me" and an AI agent fixes it for you.- Run agents on your whole github backlog and fix as many issues as possible!.
1
5
20
Interesting use case for OpenHands, automatically curating datasets for building vision-language models 🧑🔬.
@AdjectiveAlli i gave it a HF token and told it to create an art style eval dataset, came back to this
0
2
19
Merry Christmas everyone 🎄. You too can have your own AI software engineer for Christmas:. * Download it now: * Sign up for the online version: If you already use OpenHands, what do you want in the new year?.
@allhands_ai This is a great project, great Christmas present. I'm having a lot of fun.
2
1
19
See how we implemented a security verifier in our web browsing agent, preventing possibly dangerous web actions in partnership with @InvariantLabsAI.
With (web) agents on everyone's mind, check out our latest blog post (link in thread) on browser agent safety guardrails. We replicate and defend against attacks on the @allhands_ai web agent, preventing it from generating harmful content and falling for harmful requests.
0
3
18
A new course from @weights_biases features some of our learnings on agent evaluation in OpenHands, as well as a demo of OpenHands autonomously running an experiment and doing evaluation with their new product Weave 📊.
We're excited to announce the LLM Apps: Evaluation course is now LIVE! 🚀. Created in collaboration with guest experts @DynamicWebPaige and @gneubig, this course equips you with the skills needed to build trustworthy evaluations for your GenAI apps. Ready to skill up? 👇.
4
2
15
If you're eager to try out OpenAI o1 in AI coding agents, it's already available in OpenHands: Just add it in the custom model pane. Overall impressions: pretty good! We'll follow up with evals later.
2
3
14
Announcing the release of OpenHands 0.15.2, some juicy things here 🙌. - Most importantly, customizable microagents! If your project has a specialized workflow, add it to the .openhands directory in your repo to prompt the agent.
1
0
14
On Friday we released OpenHands 0.13.0 🙌. Major changes from 0.12.0:.- Integration of web browsing w/ coding for much smoother browsing.- Multimodal support by dragging in images.- Several smaller UI/UX improvements.
1
1
13
We're looking for a Chief Architect / Principal Engineer: We're looking for an experienced engineer comfortable tackling large, loosely scoped problems, and guiding a larger team towards a solution. Come build the future of software engineering with us!.
0
4
11
Introducing OpenHands 0.14.3🙌. Exciting updates:.1. One-button click to push to github.2. Incorporation of the commit0 benchmark for building apps from scratch (thanks @wzhao_nlp -- benchmark results soon!).3. Many other smaller robustness improvements
1
3
10
Thanks @tryolabs for naming OpenHands one of the top 10 open-source AI/ML/Data projects of 2024! We started in March 2024, and we're just getting starts 🚀.
🤖 Top Python Libraries 2024: AI / ML / Data Edition. Continuing our 10th anniversary celebration with 10 cutting-edge tools reshaping the AI landscape. Here are our top 5 picks that are revolutionizing how we build AI-powered applications. Let's dive.
0
1
9
We think that 2025 is going to be the year that AI agents for software development really hit the mainstream, so come join us and build this future together 💪.
0
1
9
💥.
Openhands its pretty good man. We are on the verge of a real Software Precambrian explosion.
1
0
9
We released OpenHands 0.9.8! This week we focused on making the app more reliable/snappy: - Better task canceling/restarts.- Faster startup speeds in some cases.- Configuration of config file locations. Also, new docs!:
1
2
9
What improved in the agent? Three major things:. - We switched to Anthropic's new claude-3.5 (new) model.- We modified the agent's actions to use function calling.- We made some fixes regarding directory traversal.
1
0
8
Why is this important?. OpenHands ( is an open-source dev agent that all developers can use and improve. Devs love open source, but only when it's competitive with other options. Today, OpenHands is *both* the best option, and open and freely available.
1
1
8
That's it! We're extremely excited to have state-of-the-art open agents for development, and we're even more excited to have you join our quest to build them!.- Contribute to the open source: - Apply to work with us at @allhands_ai:
1
0
7
I used OpenHands before, how big is the difference?. It's pretty big! We use OpenHands every day in our development, and things are smoother – more issues get resolved without human intervention, bigger issues can be tackled, agents don't get stuck in circles as much.
1
0
6
We're excited to be part of this amazing cohort of startups! Let's go 🚀.
Excited to announce the first 18 startups of the Menlo's Anthology Fund with Anthropic!. Anthology Batch One. In the last 6mos, we looked through 1000s of startups to get to these. CRUD business apps will collapse in the AI era, logic will shift to AI agents.
0
0
6
Results: Claude 3.5 Sonnet led the pack with a 27% resolve rate. Surprisingly, GPT-4 and the new o1 models lagged behind. Some open-source models such as Deepseek and Lllama-405B showed impressive performance too.
1
0
6
Also, we hit a milestone of pull request number 5,000!. Thanks to ketan1741 who did the honors, and all of the 200+ community members who helped us get here!.
1
0
5
How do I use it?. - Run OpenHands locally: - Run OpenHands github actions: - Sign up for our soon-to-be-released online version:
1
0
5
What makes evaluating coding agents hard? They need to execute AI-generated code, which can (1) take significant time, and (2) cause safety issues. To execute this code safely, OpenHands sandboxes code execution in a Docker “runtime”, and the agent communicates with this runtime.
1
0
5
On the price/accuracy tradeoff, Claude and Deepseek stood out. Claude offers high performance at a modest cost, while Deepseek provides decent results at an incredibly low price point of just $0.03 per issue!
1
0
5
To solve this efficiency problem, we developed a cloud-based sandboxing solution that makes it easy to evaluate many instances in parallel. This dropped evaluation time for SWE-Bench Lite from 2 days to 1.5 hours! This greatly improves our iteration time.
1
0
4
If you’re interested in trying out this evaluation environment to evaluate new LLMs, agents, or tasks, please reach out. Join our Slack ( and ping us on the remote-runtime-limited-beta channel, or message us here!.
0
0
4
This funding will go towards improving and maintaining OpenHands (, our flagship, MIT-licensed AI software developer. It can fix bugs, make improvements, write tests, etc. We’ll improve UX, agent accuracy, and applicability to large enterprise codebases.
1
0
4
Thanks so much to all the contributors, especially the new ones: @jeevaramanathan, @KLieret, @peywalt, Vaishakh-SM, amantyagiprojects, adityasoni9998, AlexCuadron, Ethan0456 🙌.
0
0
4
Using the framework, we tested both closed-source models (like GPT-4 and Claude) and open-source models (like Llama 3.1 and Deepseek v2.5) on their ability to resolve coding issues.
1
0
4
* You can download past actions and replay in headless mode.* Better handling of multiple conversations and session management.* Better handling of long-running and interactive commands.* Better handling of downloaded github repos.* Many bug fixes.
1
0
4
@ToivoMattila @gneubig For example, write "please write and run a flask server that does X" or "please run my react code with npm run dev", and it should show up here. This is a new feature though, so please report if there are any issues!.
0
0
4
Join us in our mission!.- Download OpenHands: - Contribute to the community: - Get on our waitlist for updates: - Apply for our open positions:
0
1
4
@ToivoMattila Thanks a lot for trying OpenHands out! We're actually writing a blog post right now with some examples of the various things you can do with OpenHands. We'll try to post it by Wednesday here:
1
0
3
AI is going to revolutionize development, and we believe that this should be done open-source. Devs love open source for a reason; it’s possible to download, tinker, and understand. Also, this will be powerful and important tech, it shouldn’t be restricted to a walled garden.
1
0
3
List of related projects:. 85 streamlit/streamlit 35563.86 microsoft/DeepSpeed 35370.90 All-Hands-AI/OpenHands 34828.91 openai/gym 34750.95 hiyouga/LLaMA-Factory 33865.98 ray-project/ray 33832.99 gradio-app/gradio 33750.
0
1
2
@domoritz @charliermarsh @gneubig Yep! If you give OpenHands ( a concrete prompt telling it what you want to do it should be able to make the edits, then iteratively run the tests until it works. Happy to help you try it out if you want 😃.
0
0
2
2/2.* A new microagent to prevent timing out when running npm actions.* More robust connection to sections and multi-tab support.* Migration to React Router 7 for routing.
0
0
2
@MatthewBerman Thanks for the shout-out @MatthewBerman ! We'd love to have people try it and ping us with feedback, or join the community.
0
0
2
0
0
2
@rahulvrane Definitely! We're working on multi-session support for the online version, and saving and restarting sessions will be part of this.
0
0
2
In order to ready OpenHands for general release, we made a number of major improvements to both usability and stability. OpenHands now integrates natively with github, is able to resolve previously saved sessions etc. See the blog for details.
0
0
2
And a bonus - even the script analyzing these stats? Written entirely by OpenHands! See for yourself:
0
0
2
@drivelinekyle @DrivelineBB Glad you like it! Please feel free to reach out anytime with suggestions or feature requests. Also happy to add you to our online app beta so you can easy run multiple sessions in parallel, etc. In that case just DM your github username.
1
0
2
@meanderingexile Yeah, we do it all the time on the OpenHands repo! If you get merge conflicts, just ask OpenHands to resolve them 😀.
0
0
2
@AlexTobiasDev @gaurav_dhiman @MatthewBerman Fair enough! Feedback noted on the "way too many tokens" part.
1
0
1
@WKarsens Yeah, it's an area that we're actively working on actually! Lots of room for experimentation here.
0
0
1
@DominikSeifert4 @cognition_labs @cursor_ai Interesting idea 😃 This'll be a fun thing to try in OpenHands, and of course we welcome contributions!.
0
0
1
- View the web browser pages directly in the chat interface.- The OpenHands github integration requests review from the person who started it.- A bunch of other smaller QoL improvements.
0
0
1
@RichAC2020 @gneubig All Hands can access web sites in two ways:.1. It can read individual web pages's content so the agent can reference them in its work.2. It can write programs to scrape web sites. It sounds like you might want "2.", which is definitely possible.
0
0
1
Installation of the github action is quite easy, just:.1. Add a workflow file to your github workflows: 2. Set a github token and LLM API key as repository secrets.3. Tag an issue as "fix-me" and watch the agent work.
1
0
1
One fun fact is that a significant portion of the OpenHands resolver was written by the OpenHands resolver itself -- you can see the commit log here: It has implemented new features, fixed bugs, and added documentatioon.
1
0
1
@thoughtisdead Great, glad you liked it! Feel free to ping us on github or slack if you have any issues.
1
0
1
@call_me_Nithin Yes, it's the default agent in the github repo!. (Actually we're on CodeAct version 2.2 now, with the improvements in 2.1 plus native support for web browsing).
0
0
1