0xVox

Hermes Agent v0.16.0 - “The Surface Release” Changelog below:

Graph Engineering 把不同规模的工作流程抽象为节点，让我们能够从更完整的系统视角组织 Agent 的执行与协作。但 Agent 的内部结构、协作方式仍在快速变化。 因此，我们有必要回到更基础的 Harness 层，重新理解能力究竟如何被组织起来提高协作能力的。

I spent 48 hours with the Kimi K3 modeling code. It took: - 650 mg of caffeine (mandatory) - 40 cans of LaCroix (optional... world record (?)) - 8 papers - 6 months off my lifespan Finally grokked the entire lineage of Kimi K3 and how we got here... every single step, since 2019 GPT-2

Hermes Agent is fully supported in Raft 1.0 (Previously Slock) - Give it a try!

Kimi K3 出音味来

Raven 🐦‍⬛ just crossed 2,000 GitHub stars and shipped v0.1.5. The useful signal is not the number. It’s the direction: agents need memory, but they also need better operating loops. This release adds deep research for multi-source, cited async reports, and in-tree tracing so runtime behavior can be observed without tying Raven to one backend. A better model helps. A better harness compounds. github.com/EverMind-AI/Raven

Great to see participants learning, building, and having fun at the AI for Education Hackathon at Stanford, and getting hands-on with Raven along the way. Thank you to everyone who joined us. We’re already looking forward to the next one.

launching integrations.sh today! it's an open source catalog of every products MCP / API / CLI / GraphQL server and how to authenticate to them deep links to generate api keys, 1 click copy spec urls, it's still early but i've been loving having it

介绍一下我们这个 Benchmark，它专门用来测试 Agent 的自进化能力，非常关键。 今年我们一直在挖掘 self-improving / self-evolving 这个方向，因为我们发现 Agent self-improving 在无数场景下都有实际价值。 在现在爆火的 Loop 架构中，memory-first 的自进化更是最核心的一环，可能没有之一。 正是为了把这块做扎实，我们才专门打造了这个 Benchmark。 这个 Benchmark 也是接下来我们发布 Raven 🐦‍⬛ 的重要一环。

We just renamed Slock to Raft today. The name fits what we’re building much better: a shared foundation where many agents can coordinate, carry context, and move work forward together. There’s also a quiet nod to the Raft consensus protocol — distributed actors, shared state, reliable progress. Same vibe. Sharper metaphor.

我把自家猫变成了真的活在电脑里的 3D 猫。 只需要上传几张猫咪照片，过一会儿，它就会出现在你的桌面上。 它会在屏幕边缘散步，跳上窗口发呆。 会趴在角落睡觉。 饿了会自己跑去吃饭。 最离谱的是—— 它保留了你家猫的样子。 这是我最近做过最治愈的项目。 petwins.site

We are honored to announce that two papers from the EverMind research team have been selected for oral presentation at KDD 2026, Datasets and Benchmarks Track. Of 466 submissions to the track, only 21 papers (approximately 4.5%) received this distinction. • Evaluating Long-Horizon Memory for Multi-Party Collaborative Dialogues • MPCEval: A Benchmark for Multi-Party Conversation Generation Memory for multi-party collaboration is one of the hardest open problems in AI. Our EverMemBench benchmark spans million-token, multi-user, multi-group conversations with interleaved topics and evolving facts. The findings are striking: even with oracle evidence, today's leading memory systems achieve only 26% accuracy on multi-hop reasoning that requires attributing information across participants. We congratulate the authors and thank the community for recognizing this work. We look forward to presenting at KDD 2026. Paper: arxiv.org/pdf/2602.01313