Wh1isper

AI Engineer @ConvergeAI_X | ex @YouWareAI | Open Source Contributor

Take a deep dive, then take a leap of faith.

[Wh1isper @wh1isper_]

Let the context flow, rather than adding more agents

[Converge.AI @ConvergeAI_X]

Our engineering team explores a missing layer in conversational agents: a “subconscious” that can connect changes across threads—without merging context or taking over the conversation. https://t.co/nwG5oex7hV

[Wh1isper @wh1isper_]

@xicilion 缓存冷了

[Wh1isper @wh1isper_]

灵感来源 @ewind_dev 三个代表

https://t.co/scZ8IX9m2S

[Evan (Yifeng) Wang @ewind_dev]

https://t.co/zq3RGd2o8E

[Wh1isper @wh1isper_]

https://t.co/1iJLpddcCf

[Wh1isper @wh1isper_]

https://t.co/aojUj3wb7C

[Wh1isper @wh1isper_]

Organizations should be able to shape how they work. Agent infrastructure should not become another constraint on that freedom. https://t.co/wFFquXFB9M

[Wh1isper @wh1isper_]

Organizations should be able to shape how they work. Agent infrastructure should not become another constraint on that freedom. https://t.co/wFFquXFB9M

[STRRL.gpt @strrlthedev]

最近在体验一些纯 Vibe Coding 的项目, 有我自己主导的, 也有朋友主导我在旁边视奸的

发现了一个有趣的现象, 是复杂度快速膨胀爆炸的问题. 一旦油门踩下去, AI 就会生成各种东西, 用人完全处理不了的带宽来说服你没~事~哒~

计划写得贼好看, CI 全绿, 然后线上炸了, 整得他妈跟邪教仪式似的 笑死

[Wh1isper @wh1isper_]

@vigyso @OpenAI @thsottiaux Good work!

[Wh1isper @wh1isper_]

We now support Sign in with ChatGPT.
Thanks to @OpenAI for supporting open-source projects.

cc @thsottiaux

[Wh1isper @wh1isper_]

I've seen too many people and organizations try to build agents tailored to their needs, only to end up stitching together various open-source solutions and running into one problem after another. They lack a mature approach and a clear understanding of the challenges—especially context management, memory, and observability, where real-world experience matters.

Now we're sharing how we think about and build the harness, web UI, and managed agents. Choose the approach that works for you—use the code or self-host the service to build agents for yourself or your organization. No vendor lock-in or open-source licensing headaches, and access to state-of-the-art agent capabilities.

https://t.co/8UQHEzh83k

Give us a star and let's connect!

[Wh1isper @wh1isper_]

I've seen too many people and organizations try to build agents tailored to their needs, only to end up stitching together various open-source solutions and running into one problem after another. They lack a mature approach and a clear understanding of the challenges—especially context management, memory, and observability, where real-world experience matters.

Now we're sharing how we think about and build the harness, web UI, and managed agents. Choose the approach that works for you—use the code or self-host the service to build agents for yourself or your organization. No vendor lock-in or open-source licensing headaches, and access to state-of-the-art agent capabilities.

https://t.co/8UQHEzh83k

Give us a star and let's connect!

[Wh1isper @wh1isper_]

@xiaoxxchan @xuanwo lovely

[Wh1isper @wh1isper_]

why agents api?

个人看法：agents api不一定适合开源的项目
1. 目前用户少 其实少有所谓的迁移用户
2. 自部署的时候很多假设和额外字段、插件，又要再扩展api，等于实现了一个四不像的方言

why all harness？

个人看法：harness本身（agentic loop）并没有特别之处，反而是harness的环境假设各有不同，这对于项目来说反而添加了约束，而更不适合自部署和二次开发。比如claude code是binary，codex更倾向于app-server

tldr，我个人倾向于基于社区的、可扩展性好的agent sdk来封装自己的harness，追踪第一梯队的上下文管理方式，做好Environment在harness - service之间的抽象和联动，再把记忆之类的作为插件接入即可

check ours：https://t.co/8UQHEzh83k

[Wh1isper @wh1isper_]

允许维护者修改的本意应该就是如此

[Xuanwo @xuanwo]

稍微解释一下漩涡独裁模式：

所有的 Issues / PRs 一经创建视为已经完成对辞达的贡献，我可以不经讨论对其做任意修改，我会保留作者原始的 credits，但是可能不保留任何代码

[Converge.AI @ConvergeAI_X]

Agent demos are easy. Running them reliably is the hard part.

Meet Agent Foundation (a13n): an open-source, self-hosted foundation for agents with memory, sandboxes, computer use, and durable execution.

https://t.co/uD4wt3uQdW

[Wh1isper @wh1isper_]

I've seen too many people and organizations try to build agents tailored to their needs, only to end up stitching together various open-source solutions and running into one problem after another. They lack a mature approach and a clear understanding of the challenges—especially context management, memory, and observability, where real-world experience matters.

Now we're sharing how we think about and build the harness, web UI, and managed agents. Choose the approach that works for you—use the code or self-host the service to build agents for yourself or your organization. No vendor lock-in or open-source licensing headaches, and access to state-of-the-art agent capabilities.

https://t.co/8UQHEzh83k

Give us a star and let's connect!

[Wh1isper @wh1isper_]

@yifanxu_ephai 我们今天公开了我们的harness和managed agent，其中就支持Multi Environment： https://t.co/Q2pLrO9o4V

[Wh1isper @wh1isper_]

@jalr4ever notion的计算能力不行，还得是飞书

飞书简直是在做慈善

[Wh1isper @wh1isper_]

@maker_jackie 真信他的人不就这样被分辨出来了吗

[marv1nnnnn @marv1nnnnn1]

tell me you start ml after chatgpt without telling me

[Wh1isper @wh1isper_]

@jalr4ever ai实际上放大的是资源杠杆，只是你没有钱，只有认知与行动能力

[Wh1isper @wh1isper_]

Guess what a13n stands for