🇺🇸captain america 🇺🇸

Improved infographic 😅🫣

@AiwithYasir 😴

The entire RAG industry is about to get cooked. Researchers have built a new RAG approach that: - does not need a vector DB. - does not embed data. - involves no chunking. - performs no similarity search. It's called PageIndex. Instead of chunking your docs and stuffing them into pinecone, it builds a tree index and lets the LLM reason through it like a human reading a book. hit 98.7% on financebench. beats every vector RAG on the leaderboard. no embeddings. no chunking. no vector DB. 100% open source.

@omarsar0 That’s why skills shouldn’t touch dangerous things. You need shell/python and web hooks for that

// Skills as Verifiable Artifacts // Pay attention to this one, AI devs. If you ship agent skills, your runtime is treating signed-and-cleared skills as trusted by default. This paper argues a skill is untrusted code until it is verified. The runtime should enforce that default rather than infer trust from origin. Without skill verification, HITL has to fire on every irreversible call, which degrades into rubber-stamping at any non-trivial scale. With verification as a separate gated process, HITL fires only for what is unverified. Skills are now first-class deployment artifacts. We have decades of supply-chain lessons on what happens when trust is inferred from a signature. This paper is the right ask for SKILL.md before agent skill libraries become the next attack surface. Paper: arxiv.org/abs/2605.00424 Learn to build effective AI agents in our academy: academy.dair.ai

do you understand what just happened to your computer.. Google Chrome secretly downloaded a 4GB AI model onto your device. Without asking.. Without telling you.. It's called weights.bin. It lives deep in your system folders. It powers Gemini Nano - Google's on-device AI. And if you delete it? Chrome re-downloads it automatically. Like nothing happened. Just Google deciding your hard drive is their storage unit. At 1 billion Chrome users - that's 4 BILLION gigabytes of data pushed silently across the internet. The carbon footprint alone equals tens of thousands of cars running for a year. Check your disk right now: 📁 %LOCALAPPDATA%\Google\Chrome\User Data\OptGuideOnDeviceModel To stop it: chrome://flags → disable Optimization Guide On Device Model → restart Chrome → delete the folder. Reshare so people know what's sitting on their computers.

That’s why I built ContextLattice, an infrastructure platform to run as a co-application to all your other agents. The app handles memory, context, and multi agent orchestration and communication. Features sync and async retrieval lanes for larger knowledge domains, with notifications, and back-pressure handling for writes up to 100 messages/second. Written in Go/Rust, it features novel and multimodal performant databases w many features, ready to plug into any agent Just launched premium version with hypertuned databases and experimental databases and features built in for new capabilities Memory, context, and orchestration is not a job for lil markdown files. They are great, don’t get me wrong, but I’m building for scale GitHub.com/sheawinkler/co…

// HeavySkill // One of the cleaner takes on agentic harness design I've read. They argue that what actually drives agent harness performance is not the orchestration code. It's a single inner skill: parallel reasoning followed by deliberation. If you can internalize that into the model and most of the scaffolding becomes optional. The paper systematizes this as a two-stage pipeline you can run beneath any harness, then trains it as a learnable skill via RLVR. The numbers: > GPT-OSS-20B jumps from 69.7% (M@K) to 85.5% (HM@4) on LiveCodeBench under the heavy-thinking variant. > R1-Distill-Qwen-32B nearly doubles on IFEval, from 35.7% to 69.3%. > Several models reach Pass@N-level performance with HeavySkill. Harness wins start to look like model wins once you can train them in. If parallel-reasoning-plus-deliberation really is the inner skill, the long arc is models that come with it baked in, not orchestration glue around them. Paper: arxiv.org/abs/2605.02396 Learn to build effective AI agents in our academy: academy.dair.ai

@outsource_ Welcome to Solana lol

I have no idea what’s happening but apparently someone launched a $HERMESWORLD token?? I just found out creator fees were routed to my wallet and claimed ~$500 ?!?!! Can someone explain what’s going on 😂 pump.fun/coin/2YF1qxgYV…

@rileybrown Just tell it to add the skill in the context of objectives

when /goal in codex app?

@0xSero I found warp super resource intensive when I was using it in December and January.

Warp - High level harness harness Droid - Main harness Pi - Local model harness Deepseek-V4-Pro - Main dawg Deepseek-V4-Flash - Local dawg GPT-5.5 - Backup dawg

@romainhuet My codex has superpowers because I run a grown-up memory backend. Works very well with codex and any other agent written in Go/Rust for performance as volume, lineage, and trajectories grow GitHub.com/sheawinkler/co…

We’re thinking about the next wave of Codex plugins. What’s one you’re missing today?

@regent0x_ “shipped 3 prs on the couch” is child’s play

quit his job 4 months ago to build a saas alone launched last week: $12k MRR in 6 days his entire engineering team: “2 devices online” one M4 mac mini, one main mac, both running AI agents controlled from an ipad on the couch shipped 3 PRs yesterday while watching netflix the setup looks like a joke until you see the output white desk, mac mini sitting next to a speaker, ipad with keyboard in front on the ipad screen: “WORKBENCH - 2 devices online” two purple cards showing “M4 Rex” and “Main Mac”, both connected from Minneapolis he controls everything from the couch doesn’t even sit at a desk anymore here’s how it works: the M4 mac mini runs the “builder” agent - writes features, implements specs, handles all the heavy coding the main mac runs the “guardian” agent - reviews every PR, runs tests, checks for security issues, blocks anything that doesn’t pass they communicate through a shared queue builder pushes code > guardian reviews > builder fixes feedback > guardian approves > auto-merge he watches the whole thing happen from his ipad while half-paying attention to netflix the terminal on screen shows “astrobot” running _ some kind of orchestration layer managing both agents, routing tasks, handling webhooks another window shows OpenClaw interface with chat logs: “Amazon: 2026-04-06” “Shopify: 2026-04-07” “can you launch the python inventory scheduler?” agents responding with tool outputs, status updates this is his entire product development workflow the timeline: > week 1-2: set up the dual-agent system, wrote CLAUDE.md files for each machine > week 3-4: built the MVP while agents handled 70% of the code > month 2-3: iterated based on beta feedback, agents shipped fixes overnight > month 4: launched publicly traditional estimate for this saas with one developer: 8-12 months he did it in 4, while spending half his days on the couch the ROI math: > 2 mac devices: ~$2,000 total (he already owned one) > claude subscriptions: $40/month > his time: maybe 4 hours of real work per day > revenue after 6 days: $12k MRR $144k ARR run rate from a couch and two mac minis his friends thought he was crazy for quitting “how are you going to build a whole product alone” he built a product AND an engineering team just not the kind anyone expected

@so_ainsight Yo, if you wanna parse harder, check out my complete feature and function parity rewrite of Hermes with even more intelligence builtin. GitHub.com/sheawinkler/he…

セッションをまたいで記憶を保ち、経験から自分でスキルを作って育つAIエージェント Hermes Agent。ガチで使い倒すためのGitHubレポ10選： 21 Hermes Agent 本体 Nous Research公式のコアリポジトリ。MITライセンスで自由に使える。 github.com/NousResearch/h… 2. Hermes-Wiki Hermes Agentのソースコードを解説するコミュニティWiki。実装理解に効く。 github.com/cclank/Hermes-… 3. Atlas エコシステム地図。100+のツール・スキルを俯瞰でき、RAG検索にも対応。 github.com/ksimback/herme… 4. Control Interface セルフホスト型ダッシュボード。複数エージェント・長時間タスク・記憶を一画面で管理。 github.com/xaspx/hermes-c… 5. Skill Factory タスクを振り返って新スキルを自動生成・追加。エージェントが自分の武器を自製。 github.com/Romanescu11/he… 6. Maestro ローカル動作のマルチエージェント協調ツール。Codex・Claude Code・Geminiを横断して構造化記憶と引き継ぎを管理。 github.com/ReinaMacCredy/… 7. Hermes Agent Camel 信頼境界（CaMeL）を組み込んだフォーク版。本番運用の防護に向く。 github.com/nativ3ai/herme… 8. Hermes HUD TextualベースのTUI監視ターミナル。意識の流れとメモリ状態をリアルタイム可視化。 github.com/joeynyc/hermes… 9. Hermes Alpha クラウド環境向けのHermes Agentデプロイ用テンプレート。Makefileと設定例同梱。 github.com/kaminocorp/her… 10. Awesome Hermes Agent コミュニティ厳選のプラグイン・プロンプト・教材まとめリスト。 github.com/0xNyk/awesome-… 保存して順に試してみて。

@mr_r0b0t @NousResearch yo figured you may be interested in my complete feature and function parity Rust rewrite of Hermes GitHub.com/sheawinkler/he…

For all the people who were curious about running 96+ deepseek-v4 powered @NousResearch Hermes agents in parallel, I packaged that skill up and have open sourced it on GitHub. Hope you enjoy DeepSwarm!

@thsottiaux ContextLattice as your memory, context, and orchestration infra. Written in go/rust + small llm on backend. Finally something that runs on your machine and maintains between any agents you’re running Also enable experimental features on codex! GitHub.com/sheawinkler/co…

What are we obviously not getting right with Codex?

@LLMJunky Pets incorporated to my feature and functional parity-complete Rust rewrite of Hermes Agent github.com/sheawinkler/he…

Codex App for Linux: Petsmart Edition Available on Github. Pets work on CachyOS + Wayland. Needs testing on other platforms Enjoy! Links in the comments.

@RoundtableSpace Newly integrated into my Rust rewrite of Hermes Agent. Add your favorite pet to hermes-ultra today. github.com/sheawinkler/he…

Someone built Petdex - a public gallery to browse and install Codex pets with one curl command. The AI tooling ecosystem now has its own creature collection.

@OpenAIDevs Newly integrated into my Rust rewrite of Hermes Agent. Add your favorite pet to hermes-ultra today. GitHub.com/sheawinkler/he…

Pets. Now in Codex. Use /pet to wake your pet.

@wsl8297 That’s why I built a grown up backend for agents GitHub.com/sheawinkler/co…

现在大家都在做 Agent，但一上复杂业务：架构怎么搭、记忆怎么管、多智能体怎么配合，往往越做越迷糊。 我最近看到一本开源书 Agentic Design Patterns，把智能体设计模式从入门讲到企业级，脉络清晰、拆解到位。 全书 21 章 + 7 个附录，按难度分四大部分；每章配套 Jupyter Notebook，边读边跑，理论和代码紧贴在一起。 GitHub：github.com/evoiz/Agentic-… 前半段打底：提示链、路由、并行、反思、工具调用、多智能体协作等核心模式，一次讲透“怎么设计”。 后半段落地：记忆管理、异常恢复、人机协作、安全护栏、性能评估等生产必修课，直接对准“怎么上线”。 附录还补齐框架对比和高级提示技巧，适合查漏补缺、随用随翻。 想把 Agent 从概念学到可落地的系统方法，这本开源书值得收藏，慢慢啃。

@Suryanshti777 Direct whatever agent to contextlattice instruction files and never worry about memory again. After initial setup it can auto launch on your machine upon startup and run in the background. Agent handles it from there. Human facing interfaces if desired as well

@Suryanshti777 I used some similar notions but in Go/Rust so it stays performant over time. GitHub.com/sheawinkler/co…

Someone just built this and it’s kinda insane a tool that: • spins up its own vector DB • runs embeddings locally • indexes your entire codebase • works with Claude, Cursor, Copilot and you don’t configure anything no API keys no setup no infra you just run: npx -y socraticode and it handles the rest this is how dev tools should work

@unipumpwizards yo i think this feature is broken as of past 30-40 mins

You do NOT need to connect your wallet to see your $UNIPUMP Pill Wizards. 9KrdYnHHrsWYpWMxyzbmmrjcXRr7ERWpB3byNeb1pump NFTs? No. Pump-Fungible Tokens? Yes. PFTs are the Pill Wizard from $UNIPUMP You can simply search your address to see all you have. unipump.live/my-pfts All entirely generated from code. No duplicates. Ever.

Improved infographic 😅🫣

ContextLattice is the control plane for agent memory and coordination