Brenno Ferrari

I built an Obsidian vault that makes Claude Code remember everything. You talk. It files. "Just had a 1:1 with Sarah. She's happy with auth but wants error monitoring before release. Tom said cache migration deferred to Q2." → Creates the 1:1 note → Updates the person file → Logs the decision record → Adds the win to your brag doc → Updates the active project 5 lifecycle hooks classify every message automatically. 15 slash commands. 9 subagents for heavy operations. QMD semantic search so retrieval scales. You never organize anything. The vault organizes itself. 180+ stars in a week. MIT licensed. github.com/breferrari/obs… #ObsidianMD #ClaudeCode #buildinpublic #opensource

@chebyte Glad to hear it, @chebyte! Thanks for trying it!

Been using obsidian-mind for the last few days and honestly… it’s been super useful. It’s a clean Obsidian vault template built for Claude Code that gives you a real persistent LLM wiki (same idea that’s trending with Karpathy). Claude now: → auto-organizes notes → structures projects → tracks decisions → connects ideas across sessions No more scattered context. Everything builds on top of itself. If you’re a dev running Claude Code + Obsidian, it’s definitely worth checking out: → github.com/breferrari/obs…

@RoundtableSpace We now have support to other CLIs, as requested.

Someone gave Claude Code a permanent memory using an Obsidian vault. Every session it reads your active projects, recent decisions, and past context before saying a word. The knowledge compounds instead of resetting. Claude Code is powerful but it forgets. This fixes that.

@RoundtableSpace Thanks for sharing. :)

@honestan0n @simplifyinAI Have you checked it? There's a section in the new readme explaining usage.

@simplifyinAI This doesn’t solve token usage which is the only important constraint

You can now give Claude Code infinite memory 🤯 Someone just open-sourced Obsidian Mind to give Claude Code a persistent memory backend using Obsidian. 100% Open Source.

The biggest point isn't to just not forget, or to always remember. The point is being able to take action on top of things, with an agent that is fully aware of the whole context every time. It helps you define and handle so many things... It has helped me, at least, at the personal level.

@simplifyinAI You know, all those people who wished they could forget certain things in their lives? Remembering isn't always golden, be careful.

@csaba_kissi It's a fair trade-off when under pressure to ship fast. But over time it builds debt, and then you get issues. Balance lives in-between.

Unpopular opinion: Clean code matters less than shipped code when you're starting out. Agree? 👇

@GithubProjects github.com/breferrari

-▬▬.◙.▬▬‐ ▂▄▄▓▄▄▂ ◢◤ █▀▀████▄▄▄◢◤ █▄ █ █▄ ███▀▀▀▀▀▀╬ ◥█████◤ ══╩══╩══ ╬═╬ ╬═╬ Just dropped down to say ╬═╬ Share Your GitHub Profile ╬═╬ And Let's Connect ╬═╬ ╬═╬ ☻/ ╬═╬/▌ ╬═╬/ \

@lucas_flatwhite It's good to see them giving native platform support to this.

Claude "The Advisor Strategy" = 저렴하게 똑똑하게 자, Sonnet/Haiku가 주도하고 어려운 결정에서만 Opus에 자문하게 됩니다. Opus가 좀 비싸긴 하죠. 그런데 Anthropic에서 판단했을 땐, 사실 많은 경우에 Opus 수준의 판단력이 필요한 순간은 전체 작업 중 일부에 불과했습니다. 대부분의 스텝은 Sonnet이나 Haiku로 충분한데, 그 일부 때문에 Opus를 내내 쓰는 건 낭비일 수 있죠. ↓ Anthropic이 이 문제에 대한 답을 가져왔습니다. The Advisor Strategy = Sonnet/Haiku가 작업을 주도하다가 어려운 결정에 부딪히면, 그 순간만 Opus에게 묻는 구조 에이전트가 막히는 지점에서만 Opus가 개입하고, 나머지는 저렴한 모델이 처리합니다. 단일 API 요청 안에서 이 모든 게 일어나요. ↓ 성능 수치가 매우 인상적입니다! Sonnet 단독 vs Sonnet + Opus 어드바이저 → SWE-bench Multilingual 기준 2.7%p 성능 향상, 비용 11.9% 절감 Haiku + Opus 어드바이저 → BrowseComp 기준 19.7% → 41.2%로 두 배 이상 성능 상승 비용은 줄었는데 성능은 오릅니다. 이게 중요한거죠! ↓ 왜 비용이 줄냐면.. Opus가 어드바이저 역할을 할 때는 도구를 직접 호출하지 않아요. 계획을 세우거나 수정 방향만 알려주기 때문에 보통 400~700 토큰 정도만 쓰거든요. Opus를 내내 돌리는 것과는 비교도 안 되게 가볍죠. 각 모델의 토큰은 해당 모델 가격으로만 청구되니, 실제 비용 구조도 투명합니다. ↓ 저는 이 패턴이 에이전트 설계의 기본값이 되지 않을까 생각해요. Claude Code 에서도 계속 등장하던 패턴이었죠. Opus를 쓸지, Sonnet을 쓸지.... 고민을 줄이고,, 이제는 Sonnet이 기본이고, 필요할 때 Opus를 쓴다는 방식이 자연스러워질 것 같아요. 이후 모델이 나와도 말이죠. 비용 때문에 Opus를 포기했던 사람들에게 사실상 Opus 수준을 제공하는 셈. 에이전트가 복잡해질수록 이 패턴의 가치는 더 커질거라 생각합니닷!

Essa onda de Claude+Obsidian tá interessante. Curiosamente, construí exatamente isso por razões pessoais e usei diáriamente por 1 mês até achar interessante o suficiente pra compartilhar. E daí viralizou... É um vault open-source que dá memória persistente pro agente. Cada decisão, win, pessoa, contexto, e tudo fica num grafo que o Claude lê toda sessão. v4 saiu hoje, agora roda em Claude Code, Codex CLI e Gemini CLI também. Se quiser dar uma olhadinha: github.com/breferrari/obs…

x.com/cyrilXBT/statu… EM VEZ DE ASSISTIR NETFLIX ESTA NOITE. Gaste 1 hora construindo isso. Obsidian + Claude Code = seu próprio JARVIS pessoal. Um segundo cérebro que captura tudo, conecta cada ideia e pensa ao seu lado usando o modelo de IA mais poderoso disponível. Leva 1 hora para configurar. Funciona enquanto você dorme. As pessoas que construírem isso esta noite nunca mais trabalharão da mesma forma. As pessoas que pularem isso ainda estarão tomando notas espalhadas e perdendo suas melhores ideias no próximo ano, se perguntando por que não conseguem pensar claramente. A escolha é sua.

@BHolmesDev Warp ftw

@BHolmesDev This is interesting. Will check!

The Obsidian CLI is cool, but have you heard of the `ob` CLI? 🤔 It's a "headless" version that just handles syncing, which makes it perfect for automations. Used it to have an agent fill in missing tags/categories on my notes every evening. So useful!

obsidian-mind is an Obsidian vault template that gives AI coding agents persistent memory. Hooks, commands, classification, session context. Built with Claude Code, now works with three. github.com/breferrari/obs…

obsidian-mind v4. Same vault, three agents. Claude Code, Codex CLI, and Gemini CLI now share the same lifecycle hooks. Same Python scripts. Same commands. Not a wrapper. Not an adapter. Hit Claude's rate limit mid-session. Switch to Gemini. Your vault still classifies your messages, validates your writes, injects session context. Your memory follows you across agents. 1,544 stars. Zero open issues.

@_vmlops Thanks for sharing

SOMEONE TURNED OBSIDIAN INTO A SECOND BRAIN THAT REMEMBERS EVERYTHING ACROSS AI SESSIONS Every session starts fresh. your ai forgets your goals, your decisions, your patterns you re-explain everything every single time obsidian-mind fixes that a vault template that gives your ai persistent context your north star goals, past decisions, brag doc, competency tracking all loaded automatically every session github.com/breferrari/obs… no wasted turns. no re-explaining. just deep work from minute one your notes compound. your ai gets smarter about you over time

@gaperton @simplifyinAI appreciate you, Vlad

@simplifyinAI It's not 'someone'; it's Brenno Ferrari (breferrari). What a ridiculous way to avoid crediting the authors.

@simplifyinAI thanks for the feature. I'm the someone👋

@jonoringer Nice! I've been playing with Hermes too. Will give this a try.

I built 'hermes-top' - a top-style CLI for watching Hermes activity from it's own state db. Repo: github.com/jonoringer/her… I wanted something i could monitor while my agent was working: - live session activity - recent events + history - machine load + GPU load - terminal-native dashboard It’s open source , and I’d love feedback on what would make it genuinely useful in real workflows. If you try it, let me know what’s missing, confusing, or surprisingly useful. Thanks to @NousResearch for creating Hermes.

@xoaanya Can you really call yourself a developer if you can't code with it?

Can you really call yourself a developer if you can't code without AI assistance?

@Prathkum Don't think you get very good results with just prompt and no plan, though. Plan and design just became very good prompts and decisions in-between.

Software development lifecycle: Before: Plan → Design → Build → Test → Ship → Maintain Now: Prompt → Prompt → Prompt → prompt → Prompt

@manojdotdev Claude, Grok and ChatGPT.

be honest, which one do you use the most ?