←
CAMEL-AI.org1,500 posts
CAMEL-AI.org avatar

CAMEL-AI.org

@CamelAIOrg

camel-ai.org is working on finding the scaling laws of agents. The first and the best multi-agent framework. Discord: discord.camel-ai.org. Product @Eigent_AI

◷ Joined June 2023▧ 584 Photos and videos
CAMEL-AI.org avatar
CAMEL-AI.org@CamelAIOrg·Feb 5

Open AI and Anthropic just release GPT-5.3-Codex and Opus 4.6 model, terminal capability is now on top of their list evaluating modal capability. But terminal training hits a wall fast: there aren’t enough high-quality environments. In SETA, we just shipped 1,376 validated terminal environments across: SE • sysadmin • security • debugging • networking • DevOps Compatible with Terminal Bench & Harbor. @Mike_A_Merrill @alexgshaw And we’re scaling fast 👀 Find it in: github.com/camel-ai/seta-env or search for seta-env in on harbor registry

CAMEL-AI.org @CamelAIOrg Jan 9 Article SETA: Scaling Environments for Terminal Agents — Designing resilient toolkits and scalable RL environments for CAMEL terminal agents Authors: Qijia Shen (@qijia_shen) 1,2, Jay Rainton 3, Aznaur Aliev 4, Ahmed Awelkair (@ahmedawelkair) 1,2,4,5,

Guohao Li 🐫 avatar
⟳ CAMEL-AI.org retweeted
Guohao Li 🐫@guohao_li·Nov 11

Can agents perform agentic RL? If RSI happens, agents will need to take over researchers' work on agentic RL Introducing AgenticRL-Bench: the first release from The Environment Lab @env_lab_ai effort at Eigent AI 22 tasks rebuilt from real-world RL incidents on production-grade agentic RL systems, including Miles, Slime, Verl, SGLang, and Harbor, spanning 8 areas of the stack: training data, harnesses, algorithms, RL infrastructure, recipes, evaluation, monitoring, and capability transfer We ran 6 agents with 3 attempts per task. On the 19 verifiable tasks, no agent passes more than 14.0% of its attempts Across the three open-ended tasks, agents improved real policies or training steps using 4×H200 GPUs over 10–16 hours. On one of these tasks, Zero Signal Hill Climb, where the base model fails to solve any of the 38 tasks, the best attempt developed its own method: the agent first used hinted text feedback to generate successful rollouts, then removed the feedback from the trajectories and blended these successful rollouts with regular GRPO rollouts to form mixed training groups. It monitored the training process and adjusted KL loss weight to prevent training collapse. This improved pass@1 from 0 to 0.186 (12 out of 38 tasks)

Post attachment
Guohao Li 🐫 avatar
⟳ CAMEL-AI.org retweeted
Guohao Li 🐫@guohao_li·Nov 11

try gemini 4 carbon on eigent - open source cowork agent desktop agent day 0 support as always 🫡

Gemini 4 Carbon is now on Eigent!! Given only raw lab instrument files, it rebuilt a published drug-discovery pKa dataset from scratch in one prompt! 👉 reads an undocumented instrument format -> fits spectra & extrapolates methanol-water data back to pure water -> decides which measurements to trust, like a lab scientist would Try Gemini 4 Carbon on Eigent via byok or cloud api today!

1:38
Zecheng Zhang avatar
⟳ CAMEL-AI.org retweeted
Zecheng Zhang@zechengzh·Oct 20

We're releasing AgentVFS, a virtual filesystem interface for AI agents agentvfs.org/ AgentVFS unifies every service into one interface and replaces the need of calling MCP. Stored data becomes one virtual filesystem across services, described by VFS.md, and actions become virtual CLIs, so the agent no longer calls MCP tools at all. AI agents work on live data across every service through the terminal-like interface they know best. They use the Unix tools they learned in training (ls, grep, jq, pipes, loops) and the terminal workflows they practiced in post-training on agentic coding. We're also releasing VFS-Bench 1.0 agentvfs.org/benchmark/vfs-b…, the first systematic study and benchmark to compare a virtual filesystem with MCP on the same tasks. Compared with MCP, AgentVFS uses up to 88% fewer tokens and has about a 5% higher success rate on average.

Post attachmentPost attachment
0:09
Guohao Li 🐫 avatar
⟳ CAMEL-AI.org retweeted
Guohao Li 🐫@guohao_li·Oct 20

What if every SaaS service could be virtualized as files? Excited to release AgentVFS and VFS-Bench 1.0 with @zechengzh: a virtual filesystem interface for AI agents, and the first systematic benchmark comparing it against MCP on the exact same tasks One task, two interfaces: VFS v.s. MCP. 200 tasks from MCP-Atlas and Toolathlon. Every task runs once through MCP tools and once through AgentVFS, where services show up as files described by VFS.md and actions become CLIs What we found: - Claude Opus 5.5: 86% vs 83% pass@1, with 166k vs 617k tokens per task (73% fewer) - GPT-6.1 Sol: same 75% success, 72% fewer tokens - GPT-5.6 Luna: 61% → 73% success, 60% fewer tokens - Up to 88% fewer tokens, and about 5% higher success on average The part I find most interesting: stronger models gain the most. Opus and Sol save over 70% of their tokens, while smaller models save much less. Frontier models were post-trained heavily on terminal-style coding, and they seem to bring all of it to ls, grep, jq and pipes once you give them that interface We spend all day building RL environments for agents. One lesson keeps coming back: the interface is part of the environment. The same task can get much easier or much harder depending on how you expose the world to the model An AgentVFS layer may be how we scale environments for agents. One filesystem in front of every service, and the terminal skills models already have carry over to all of them

Post attachment

We're releasing AgentVFS, a virtual filesystem interface for AI agents agentvfs.org/ AgentVFS unifies every service into one interface and replaces the need of calling MCP. Stored data becomes one virtual filesystem across services, described by VFS.md, and actions become virtual CLIs, so the agent no longer calls MCP tools at all. AI agents work on live data across every service through the terminal-like interface they know best. They use the Unix tools they learned in training (ls, grep, jq, pipes, loops) and the terminal workflows they practiced in post-training on agentic coding. We're also releasing VFS-Bench 1.0 agentvfs.org/benchmark/vfs-b…, the first systematic study and benchmark to compare a virtual filesystem with MCP on the same tasks. Compared with MCP, AgentVFS uses up to 88% fewer tokens and has about a 5% higher success rate on average.

Post attachmentPost attachment
0:09
Eigent AI avatar
⟳ CAMEL-AI.org retweeted
Eigent AI@Eigent_AI·Oct 19

Eigent v1.0.6 is live! Models, Automations, and Session recovery just got a major upgrade 🚀 What’s new: 🧠 Manage saved models in one place ⏰ Inspect Automation schedules and run history 👋 Choose your Work Profile 🔄 Recover Sessions and handle approvals more reliably 📂 Preserve drafts, terminal history, and finished files 📦 Validate Bundles before sharing and recover failed imports 🔁 Follow clearer upgrade guidance ❤️ Enjoy more control over everyday work Update to v1.0.6, explore the new features and share your feedback with us 🙌 Read the full release notes: eigent.ai/blog/eigent-v1-0-6…

Guohao Li 🐫 avatar
⟳ CAMEL-AI.org retweeted
Guohao Li 🐫@guohao_li·Oct 17

@Eigent_AI is proud to be part of the SGLang Summit! Come talk to us about RL × Environments × Agents!

Proud to welcome our newest sponsors to SGLang Summit. In just one week since our public announcement, our sponsor lineup has grown from 11 to 36. Thank you to every sponsor helping bring this community together. Join 2,000+ engineers, researchers, and builders from across the AI ecosystem for technical sessions, hands-on workshops, and the GPU Expo. Hear from Lip-Bu Tan, Lilian Weng, Soumith Chintala, Aravind Srinivas, and more. Meet the teams building across AI hardware, infrastructure, models, and applications. 🗓️ November 12–13, 2026 🌉 Fort Mason, San Francisco 🎟️ Get your tickets: sglang.io/summit

Post attachment
SGLang avatar
⟳ CAMEL-AI.org retweeted
SGLang@sgl_project·Oct 16

Proud to welcome our newest sponsors to SGLang Summit. In just one week since our public announcement, our sponsor lineup has grown from 11 to 36. Thank you to every sponsor helping bring this community together. Join 2,000+ engineers, researchers, and builders from across the AI ecosystem for technical sessions, hands-on workshops, and the GPU Expo. Hear from Lip-Bu Tan, Lilian Weng, Soumith Chintala, Aravind Srinivas, and more. Meet the teams building across AI hardware, infrastructure, models, and applications. 🗓️ November 12–13, 2026 🌉 Fort Mason, San Francisco 🎟️ Get your tickets: sglang.io/summit

Post attachment
CAMEL-AI.org avatar
CAMEL-AI.org@CamelAIOrg·Oct 16

We're starting in 10 minutes! Join us for Zihan @CrD104 's Live Talk on CAD World. 🔗 Zoom: us06web.zoom.us/j/8190844404…

📢 Join our CAMEL-AI Live Talk next Friday with Zihan Dong @CrD104 , AI Solutions Engineer at Lenovo, presenting CADWorld: Computer-Use Benchmark for Long-Horizon Computer-Aided Design He'll introduce CADWorld, a benchmark that evaluates how effectively AI agents can perform realistic mechanical CAD workflows by operating professional software directly through its GUI. It includes 200 tasks across 11 workflow categories, covering the end-to-end design process from modeling and assembly to manufacturing and simulation. 📅 October 16, 2026 🕐 14:00 BST | 9:00 EDT | 21:00 CST 🔗 Register: docs.google.com/forms/d/1Ow1… 🐫 If you're interested in computer-use agents, CAD, or AI for engineering, don't miss this talk!

Post attachment
Eigent AI avatar
⟳ CAMEL-AI.org retweeted
Eigent AI@Eigent_AI·Oct 14

Gemini 4 directed a multi-shot 3D Blender scene on Eigent! 🎬 From a calm setup to a delicate wind-blown flight and final landing. 4 connected shots, dynamic DOF, and consistent physics constraints in just one prompt! Build effortless 3D creations today on Eigent via byok & cloud API!

1:05

Today we are rolling out Gemini 4 Argon globally in the Gemini API, to Pro + Ultra in Gemini App, and Ultra subs in Antigravity! Argon is a huge step forward for Gemini, and a labor of love. This model has been battle tested by over 100,000 people at Google over the last month, with so much new innovation, ambition, and hard work baked into this release. The team has been heads down over the last 6 months working to make this possible across the stack (research, infra, safety, eng, etc). We rolled the model out to cyber defenders as part of our fairwind program and early access customers over the last 2 weeks, it has been amazing to see and talk to folks who are now using this model daily. Seeing the models writing capability, raw knowledge and persistence for coding, paired with new features like 1M token output when the model needs to work more has gotten me even more excited about our path ahead. This is just the start of the Gemini 4 era. As you saw with 3.5, 3.6, 3.7, and 3.8 Flash, our iteration speed with Gemini is fast : ) Now with Gemini 4 Argon to work from, I am cannot wait to see the progress continue, and most importantly see what this new model enables for the world. This is such a special moment in the arc of technology, as millions of people everyday tackle problems they could never have thought possible before. Gemini 4 Argon was built to help you tackle the hardest problems. I have personally felt this spark with Argon and hope you do too. Onwards!!

Eigent AI avatar
⟳ CAMEL-AI.org retweeted
Eigent AI@Eigent_AI·Oct 13

Gemini 4 Argon is now on Eigent!! It turns terminal bench author's insights into agent skills in one prompt! 👉 understands RL videos -> extracts key insights with precise timestamps -> produces validated agent skills plus a structured HTML report Add your Gemini API key on Eigent and try it today!

1:17

Today we are rolling out Gemini 4 Argon globally in the Gemini API, to Pro + Ultra in Gemini App, and Ultra subs in Antigravity! Argon is a huge step forward for Gemini, and a labor of love. This model has been battle tested by over 100,000 people at Google over the last month, with so much new innovation, ambition, and hard work baked into this release. The team has been heads down over the last 6 months working to make this possible across the stack (research, infra, safety, eng, etc). We rolled the model out to cyber defenders as part of our fairwind program and early access customers over the last 2 weeks, it has been amazing to see and talk to folks who are now using this model daily. Seeing the models writing capability, raw knowledge and persistence for coding, paired with new features like 1M token output when the model needs to work more has gotten me even more excited about our path ahead. This is just the start of the Gemini 4 era. As you saw with 3.5, 3.6, 3.7, and 3.8 Flash, our iteration speed with Gemini is fast : ) Now with Gemini 4 Argon to work from, I am cannot wait to see the progress continue, and most importantly see what this new model enables for the world. This is such a special moment in the arc of technology, as millions of people everyday tackle problems they could never have thought possible before. Gemini 4 Argon was built to help you tackle the hardest problems. I have personally felt this spark with Argon and hope you do too. Onwards!!

CAMEL-AI.org avatar
CAMEL-AI.org@CamelAIOrg·Oct 9

📢 Join our CAMEL-AI Live Talk next Friday with Zihan Dong @CrD104 , AI Solutions Engineer at Lenovo, presenting CADWorld: Computer-Use Benchmark for Long-Horizon Computer-Aided Design He'll introduce CADWorld, a benchmark that evaluates how effectively AI agents can perform realistic mechanical CAD workflows by operating professional software directly through its GUI. It includes 200 tasks across 11 workflow categories, covering the end-to-end design process from modeling and assembly to manufacturing and simulation. 📅 October 16, 2026 🕐 14:00 BST | 9:00 EDT | 21:00 CST 🔗 Register: docs.google.com/forms/d/1Ow1… 🐫 If you're interested in computer-use agents, CAD, or AI for engineering, don't miss this talk!

Post attachment
CAMEL-AI.org avatar
CAMEL-AI.org@CamelAIOrg·Oct 3

Thank everyone who contributed to SETA! Check it out if you are working on terminal agent RL: github.com/camel-ai/seta

SETA got accepted to NeurIPS 2026! 🎉 We released the full training scripts for DeepSeek-V4-Flash, GLM-5.2 (LoRA), Inkling-Small, and Qwen3.8-27B, with both GRPO and PPO, alongside 4,500+ open-source, verifiable terminal RL environments Huge thanks to the @radixark Miles team for helping us scale the experiments! github.com/camel-ai/seta/tre…

Post attachment
Guohao Li 🐫 avatar
⟳ CAMEL-AI.org retweeted
Guohao Li 🐫@guohao_li·Oct 2

SETA got accepted to NeurIPS 2026! 🎉 We released the full training scripts for DeepSeek-V4-Flash, GLM-5.2 (LoRA), Inkling-Small, and Qwen3.8-27B, with both GRPO and PPO, alongside 4,500+ open-source, verifiable terminal RL environments Huge thanks to the @radixark Miles team for helping us scale the experiments! github.com/camel-ai/seta/tre…

Post attachment

🎉 SETA: Scaling Environments for Terminal Agents has been accepted to the #NeurIPS2026 Evaluations & Datasets Track! We release SETA-Env, the largest open-source verifiable terminal RL dataset (4,500+ envs), with full synthesis & training pipelines 🧵 arxiv.org/abs/2607.10891

CAMEL-AI.org avatar
CAMEL-AI.org@CamelAIOrg·Sep 17

Thanks for highlighting our work. Great work @AtriaASI!

We’re excited to highlight SETA, an open-source RL environment for training and evaluating AI agents from the @CamelAIOrg . A valuable resource for researchers and developers exploring agentic reinforcement learning and complex task environments. Check it out and support the CAMEL-AI community! 🚀 🔗 github.com/camel-ai/seta

Eigent AI avatar
⟳ CAMEL-AI.org retweeted
Eigent AI@Eigent_AI·Sep 13

@OpenAI GPT 6 Astra is now on Eigent! ⚡️⚡️⚡️ Our design engineer @douglas_ym used Eigent with GPT 6 Astra to rebuild his old master’s project, TaSH, from just a few photos of the original prototype. Eigent recreated the full product in Blender, filled in details missing from the original prototype, and automatically rendered a 30-second @Apple style product showcase video. Download Eigent today. Give Eigent a few images from one of your forgotten projects and see what GPT 6 Astra can rebuild. Fully open source. Self-hostable on your desktop. Even for confidential projects, your data can stay completely under your control.

0:46