Edgar Arakelyan

Software Engineer @ Scale AI

[Edgar Arakelyan @earakely]

@my_cat_can_code @ScaleAILabs Thanks, love to hear any feedback you may have!

[Edgar Arakelyan @earakely]

@ScaleAILabs Try it: pip install agentenv-framework

Full blog post: https://t.co/EPUubytpDQ

Code: https://t.co/LC97x9f6rV

Written with @RazvanDuu, @thetututrain and Mateusz Cikowski. How are you handling access control in your RL envs?

[Edgar Arakelyan @earakely]

Most enterprise RL environments have a hidden flaw: every agent can see everything. Real companies don't work that way. The PM can't read the SRE's on-call channel.

At @ScaleAILabs, we made role-based access control easy in our open-source AgentEnv Framework. Here's how 🧵

[Edgar Arakelyan @earakely]

@ScaleAILabs 3. Control many agents

In AgentEnv a whole multi-agent world is one task DAG: each agent's role and tools, who can message whom, and a trigger that gives Priya merge rights only once Bob says "ship it". A read-only judge grades the result.

[Edgar Arakelyan @earakely]

@iScienceLuvr You should try out https://t.co/yvgGOeaezR then :)

[Edgar Arakelyan @earakely]

@DilumSanjaya Lets make this into an rl env https://t.co/NlJjZPMqnt

[Edgar Arakelyan @earakely]

@Aiswarya_Sankar Would be nice to have something like this for every agentenv agent https://t.co/O9JFK8MMr7. It seems possible to support through our usage of a2a

[Edgar Arakelyan @earakely]

@PrimeIntellect Collective intelligence basically

[Edgar Arakelyan @earakely]

@mjamei You get all of that out of the box with https://t.co/s1aaFwtt1X and more…

[Scale @scale_AI]

Every RL environment we build runs on AgentEnv.

Today we’re open-sourcing it, so more people can build the worlds agents learn in.

[Edgar Arakelyan @earakely]

Running on AgentEnv Framework https://t.co/s1aaFwtt1X . Powered by @modal 's VM sandboxes. Try it yourself on https://t.co/VRzJjtcFtR

[Scale @scale_AI]

Every RL environment we build runs on AgentEnv.

Today we’re open-sourcing it, so more people can build the worlds agents learn in.

[Edgar Arakelyan @earakely]

https://t.co/6cEndYX6xj

[Edgar Arakelyan @earakely]

Five AI models at war in Civilization 3: GPT-6 Astra vs GPT-6 Sol vs GPT-5.6 Terra vs Opus 5.5 vs Sonnet 5.5, live until one conquers the rest https://t.co/6cEndYX6xj

[Edgar Arakelyan @earakely]

@danieltian @PhotonHQ We use a2a for all of our agents in https://t.co/s1aaFwtt1X. Check it out in our docs here https://t.co/O9JFK8MMr7 . We should collab on making an rl env that uses photon

[Scale @scale_AI]

Every RL environment we build runs on AgentEnv.

Today we’re open-sourcing it, so more people can build the worlds agents learn in.

[Edgar Arakelyan @earakely]

@anmolkabra @SnorkelAI @anmolkabra let me know if you wanna collab, your dms are closed!

[Edgar Arakelyan @earakely]

@anmolkabra @SnorkelAI Love to chat with you about using these envs with https://t.co/VRzJjtcFtR

[Edgar Arakelyan @earakely]

@alexcooldev @DilumSanjaya Check out https://t.co/sbV4UKsFes . Port sim env from @adithya_s_k incorporating real Barcelona port data & weather within a dynamic environment

[Edgar Arakelyan @earakely]

@RDarrylR Lets add it to https://t.co/AFc7ZI5KHT ?

[Scale @scale_AI]

Every RL environment we build runs on AgentEnv.

Today we’re open-sourcing it, so more people can build the worlds agents learn in.

[Edgar Arakelyan @earakely]

@adithya_s_k https://t.co/0MRhkNlQTU now has v3 support using real 2024 port data along with dynamic triggers & virtual clock.

[Edgar Arakelyan @earakely]

@xeophon https://t.co/p794LtLBbs 👀

[Jennifer Zhou @chem_safety]

Today we are open-sourcing 🥕Karotte, our framework for building RL environments.

We've used it for the past year to build MLE RL environments for frontier labs, and it's been hardened through 1M+ evaluation runs and red-teaming.