ex @Dify_ai | Building Myrio | ACMer & Computer Science Enthusiast | Lesbian | 💖love maimai & 📝writing | 🐱Catgirl | 💝I'll always be with you.
Shibuya-ku, Tokyo
Joined August 2023
- Tweets10K
- Following541
- Followers6K
- Likes34.7K
Pinned Tweet
Know you will wake, on winds rise again
知道你终会醒来，从那之后天空即为极限
For this journey's end but one step forward to tomorrow…
而这旅途的终焉，即为迈向明天的开始
Mario's right: history tools are the way.
We built one. Obelisk is an open-source session retrieval layer for coding agents. Your agent can:
- Trace why a piece of code was written that way
- Pick up another agent's unfinished task
- Dig through months of sessions to understand how the architecture evolved
Raw sessions, queried on demand. No embeddings, no memory extraction.
And with Code Mode, agents can explore history without flooding their context.
github.com/tommy0103/obelisk
还是来正式宣传一下！昨天在 GOSIM 没讲尽兴的东西（
我们做了 Obelisk，一个让 Coding Agent 能够检索自己所有历史 Session 的开源 Infra
它可以追溯一段代码为什么这样写、接手另一个 Agent 的任务，甚至研究过去两个月的开发历史，重新梳理整个项目的架构演进
而我们没有使用任何 Embedding 或向量数据库
🔗 GitHub: github.com/tommy0103/obelisk
为什么已经有那么多 Agent Memory 产品，我们还要做一个？
因为我们觉得，传统 Memory 可能从一开始就搞错了问题：
为什么要让模型在过去，决定未来需要记住什么？
预先提取的 Memory 不仅会丢失原始语境，还可能让 Agent 把局部结论当成普遍事实
我们的思路是：与其让 Agent 记住过去的结论，不如让它有能力重新研究过去
技术上也很简单：我们放弃了 Embedding 和向量数据库，使用 BM25 / FTS + Code Mode，让 Agent 自己搜索、阅读、验证历史
实际效果比想象中有趣：从追溯代码设计、跨任务 Handoff，到生成 Roadmap，甚至复盘过去一两个月的架构决策，都可以完成
部分 Benchmark 上的表现也优于 Mem0，我们还在寻找更有说服力的评测方式
更完整的技术讨论：
obelisk.antinomie.org/blog/
欢迎来玩，也欢迎挑战我们的设计！
Hey @badlogicgames , curious what you think of Obelisk!
I agree with your take that we don't need a memory system when we use coding agent for something. But agents still need access to their history.
Documents are only lossy projections of past work, while using code to preserve history has a much larger blast radius and context cost.
We think on-demand session retrieval offers a surprisingly cheap alternative. With Code Mode, agents can compose queries and precisely control how much history enters their context.
Given Pi's new-session design, I'm curious whether you'd find a built-in session retrieval layer valuable.
(Attached a couple of slides from my GOSIM talk!)
拿到新设备第一件事情是装 lody，第二件事情是装 magpie
lody 和 magpie 真的好用，一个一键 agent setup，一个一键 ai provider setup
我需要做的所有事情就是 add agent，add provider，去 kimi 和 deepseek console 上拷贝一下 api key，也不需要记住 base url，太舒服了
本来换主力设备是一个很痛苦的过程，但至少在 vibe coding toolchain 上没那么痛苦（反而最痛苦的事情变成了如何迁移本地 workspace 和 repo
@yetone @lody_ai That's so cool!
Yuu💖 retweeted
LayerFS v1.0.0 is out 🎉 Open-source (MIT)
Recursive filesystem for AI agents.
6 core ideas:
1. Ephemeral computing: agents spend ~80% thinking, so each Bash call mounts a Workspace, then commits
2. Decentralized agents: every agent commits to one Store, no DB server
3. Multiagent coordination: every tool call is a Commit; same-file clashes refused
4. Recursive filesystem: forks copy no files, at any depth
5. Ultra storage: dedup, then compression: 10 copies → 0.31
6. Instant bootstrap: ignored files stay in every Commit. ~5 min install vs 0.05 s mount
And 4 more:
7. Auto checkpoints: fork from any step
8. Append-only objects: content-addressed, never rewritten
9. .lww: lockfiles and caches take the newest edit
10. Big file, small edit: a write keeps only its bytes
Here are our links:
github: github.com/Ephemeral-AI-Lab/…
website: layerfs.ai/
Watch the video below, and have Fun!!!
Zed 的选择再次证明在 Obelisk 中选用 bm25 / FTS5，弃用 embedding 是正确的，这些决定让 Obelisk 变得简单和足够小，且不伤害检索的性能和可靠性
Yuu💖 retweeted
Last year’s April, I visited Cloudflare and learned a lot about Cloudflare Workers. It got me thinking about what we could build on top of it.
Fast forward one year, and we’ve built ARC (Agentic Realm Computer) on top of Cloudflare’s infrastructure.
I feel incredibly lucky I made that visit. Sometimes, a single conversation can spark something much bigger than you ever imagined.
Yuu💖 retweeted
国庆的时间说动 @frostming90 做了一件在最开始做 bub 的时候我们一直想干但没有干的事情。
重写后的 republic 是一个纯粹的 AI provider sdk，没有任何 Agent 或者 Gateway 假设，只是提供一致的抽象来对接各种 Provider 。
同时提供了对 OAuth 等鉴权方式的支持，所以你可以通过一套接口自由使用你的 Codex 订阅或者 DeepSeek API 。
Introducing Republic 0.6.0.
An AI provider SDK that iss just a provider SDK.
OpenAI, Anthropic, Gemini, DeepSeek and more. Chat, streaming, tool calls, structured output. One Python API.
Built for providers, not an in-house harness or gateway.
github.com/bubbuild/republic
Really loved GOSIM Shenzhen! These 3 days were unforgettable. It was especially meaningful to share some insights on building Obelisk at the Agentic OS & App track. I am grateful for the conversations and the energy in the room.
Looking forward to seeing everyone again at GOSIM next year☺️
Two days. 2,000+ builders. Pure open-source energy.
From agentic OS architecture to real-world open-weight model deployments, GOSIM Shenzhen 2026 proved that the next wave of open tech is already here.
Huge thanks to every speaker, developer, and community contributor who packed the halls.
The conversations don’t stop here. 🚀
#GOSIMShenzhen #OpenSource #AgenticAI #DevCommunity
Want to know our speech?
xitter.cf/QuantumTransf/status/2…
还是来正式宣传一下！昨天在 GOSIM 没讲尽兴的东西（
我们做了 Obelisk，一个让 Coding Agent 能够检索自己所有历史 Session 的开源 Infra
它可以追溯一段代码为什么这样写、接手另一个 Agent 的任务，甚至研究过去两个月的开发历史，重新梳理整个项目的架构演进
而我们没有使用任何 Embedding 或向量数据库
🔗 GitHub: github.com/tommy0103/obelisk
为什么已经有那么多 Agent Memory 产品，我们还要做一个？
因为我们觉得，传统 Memory 可能从一开始就搞错了问题：
为什么要让模型在过去，决定未来需要记住什么？
预先提取的 Memory 不仅会丢失原始语境，还可能让 Agent 把局部结论当成普遍事实
我们的思路是：与其让 Agent 记住过去的结论，不如让它有能力重新研究过去
技术上也很简单：我们放弃了 Embedding 和向量数据库，使用 BM25 / FTS + Code Mode，让 Agent 自己搜索、阅读、验证历史
实际效果比想象中有趣：从追溯代码设计、跨任务 Handoff，到生成 Roadmap，甚至复盘过去一两个月的架构决策，都可以完成
部分 Benchmark 上的表现也优于 Mem0，我们还在寻找更有说服力的评测方式
更完整的技术讨论：
obelisk.antinomie.org/blog/
欢迎来玩，也欢迎挑战我们的设计！