Chakra Labs
@chakra_ai
Followers
11K
Following
312
Media
88
Statuses
203
Encoding human taste into intelligence. The open data protocol for AI and beyond.
Brooklyn, NY
Joined June 2024
Today, we’re excited to announce Dojo, a collaborative RL environment suite for computer use agents (CUA).
27
36
293
An interesting design decisions we made with dojo: - All of our dojos inherit the following state pattern
1
1
10
so we did this :)
if i was running Linear, and i wanted every AI ever to become very good at using Linear, i would make an RL environment version of Linear and open-source it for free, with a curated set of realistic sample data. every lab would train with it. who else can compete with that?
4
7
395
This is awesome... This is how we craft the future of CUAs, by actually open-sourcing the RL env suites...
Today, we’re excited to announce Dojo, a collaborative RL environment suite for computer use agents (CUA).
0
1
9
excellent work from @chakra_ai on the Dojo suite, which integrates seamlessly with verifiers + Environments Hub a full SDK for building real-world web applications as RL environments very exciting to see the open environment ecosystem growing stronger every day :)
3
5
99
To get started with Dojo, check out our website at
trydojo.ai
The RL Environment Hub for Computer Use Agents
2
1
23
We've also launched dojo-bench-mini, a small public benchmark of tasks against our high quality mocked environments. https://t.co/0JPBfmMARY
3
2
23
Integrations: Dojo has first class support with popular RL frameworks like verifiers (via @PrimeIntellect) and verl (via ByteDance Seed). Support for more frameworks coming soon.
1
1
28
Dojo brings these clones from closed off walls to the general public, so anyone can train on them and contribute towards them.
1
0
23
Historically, frontier labs have worked with data vendors to build clones of top websites internally. These clones are useful because reinforcement learning tasks can be written against them without having to worry about changes in UI, CAPTCHA, or authentication.
1
1
26
Dojo allows you to seamlessly spin up interactive environments with tasks - allowing you to train and evaluate top CUA models like Sonnet 4.5, GPT-5/4o, and self-hosted models.
2
2
30
Dojos are interactive CUA environments where models learn by doing. They can interact with elements on screen, play games, solve puzzles, perform actions on productivity software - and learn from rewards defined from these tasks.
1
1
24
Today, we're excited to announce GLADOS-1, the first computer-use (CUA) model post-trained using collective, crowd-sourced trajectories. Post-trained on UI-TARS-7B-SFT, we improved performance of the base model on OSWorld.
22
42
208
To explore Farcaster data in Chakra, check out our marketplace and start building your own social analytics workflows. https://t.co/PQLzGsb1Vl
chakra.dev
Bringing Farcaster data into Chakra by partnering with The Indexing Co
3
10
46