Isn't it beautiful to watch? Compounding.

Also relevant:

Actually make it those 3, save them, give them your AI to Analyse and index.

What’s the difference between using this Vs just writing to a db?

do you not see it?

built exactly this. obsidian as the knowledge graph, daily notes auto-ingested, semantic recall across sessions. the agent that remembers beats the agent that reasons

show us 👀

> The bridge between Claude and your vault is the MCP — model context protocol. why? Obsidian released CLI, wdyn mcp?

@shipulin It overlapped with the release of the CLI

You would like

I've wanted to learn how to use Obsidian but it's a daunting task in itself. I'm going to check this out.

I'll publish something for you soon

Have you already tried with Codex?

I do - they share the same obsidian brain

👀

Thanks for the idea. Creating a hybrid setup with my own memory system with this.

Cool - show us what you are building 🫡

I already have two Qwen3 embedding models running for two isolated agents running in the same app. You just gave me the upgrade. A “Dual-Brain” Hybrid that adds a structured, human-auditable knowledge layer on top of my existing semantic memory (SQLite + Qdrant vector indexes) without touching or replacing it. My two agents, Lexi/Aria, each get their own completely isolated “brain”: separate Markdown vaults, graph indexes, event pipelines, and retrieval policies. Everything stays local, shared code only, no passive memory bleed between them. Eventually I’ll layer that into an intelligence layer that sits above both of them and autonomously pulls from both into a single shared memory for the entire system. Should be fun when it’s finished.

Exciting times!

Memory has been the biggest challenge for me. I have 350+ hours of audio transcripts, 250+ articles, and hundreds of pages of writing. We’ll see how this does.

Wow 🤯

👨‍🚀

its beautiful but what is its practical application?

Structured and linked memory and knowledge - and the visualization 😍

Holy shit nyk lol

haha, I also love it

Been doing something similar with structured memory files that persist across sessions. The difference is night and day — instead of re-explaining your entire setup every time, the agent just picks up where it left off.

Let us see what you built 👀

Fantastic post

@le_Chateau_d_If Appreciate 🫡

Context is everything 🌌

🫡

Arguably the best agentic KOL in the game. Keep smashing it chief.

I'm just 2 weeks in

Whats your suggestion to create an persistent memory on an amateur basis for example for Gemini used as a plugin to an app? What would be the effect of saving all conversations as .md and hook it to check files on a regular basis?

@vidalnext

I should have read that before replying. I don’t know what I want to do

So today I gave my agents shared memory. And the wild part? I spun the whole thing up on my laptop in about an hour @origin_trail $TRAC

This framing finally gave me language for something I’ve been solving by feel. Building something similar for sales teams same four failure modes, just slower and harder to see than in code. Saved this one. Good work.

👨‍🚀

a functioning second brain fr

Work in progress 🤝

just built this, in the same spirit: it works SO well.

Nice, will check. Official LACP release also coming soon.

goals… 😍😍😍

🤯

🧑‍🚀

How long till this is part of Claude if it works?

i have too much personal data in obsidian that i can't trust claude with like passwords

@bl888m Fair - passwords I don’t store there - switch to 1Password for passwords and clean up the obsidian for your AI brain!

I could watch it all day

I have chromadb linked to memgraph by duplicated entry key's. Every prompt triggers a semantic search that pulls what we've already discussed from the memgraph and appends it to the current context. Chatbot+

👀👨‍🚀

How did you create the visual or did you give your agent a prompt to design the visual for the memory ?

The UI is @obsdmd, with plugins added and color-coding for categories/topics and automations. The animation you see is the re-indexing of Obsidian on startup, since the last time you used the app. I'm announcing and publishing my local agent control plane, including all this, soon. It's public already but I want to have some more additions before I go fully public with it.

@obsdmd Nice, I created my memory system based off a mushroom network lol I started with a visual brain model first to understand wtf I was doing. I think it’s dope everyone is building their own memory for their harness, especially with visuals to show the indexing. I’ll follow that @

@obsdmd That’s amazing! Would love to learn more about- the mushroom network analogy is extremly interesting!

@obsdmd I gotta organize everything I been just posting as I build but most of it is on my account. The mycelium mushroom network self organizes, and treats input Like knowledge graph coordinates. How it values the coordinates is also security. Less valued coordinates decay overtime

Aww that's really interesting! The concept of compounding can be quite fascinating, especially in the context of AI development.

It is hella fun!

🎮

👨‍🚀

Obsidian is flat files which means no structured database, no SQL queries, no semantic search.

Flat files ≠ no structure. Dataview gives you SQL-like queries over frontmatter, Smart Connections does local semantic search with embeddings, and QMD adds vector indexing with local models for retrieval. Because it's markdown, you get git versioning, zero lock-in, and any AI agent can read/write your vault via MCP. The filesystem is the database.

Is this autonomously growing?

Semi-autonomous. Pipelines ingest sessions, transcribe voice/video, synthesize knowledge, and index vectors automatically. But writes to the knowledge graph are gated — dry-run by default, human approves. It grows on its own but doesn't overwrite without consent.

Very cool! Thanks a million!

Unofficial alpha:

8 months ahead as usual. Follow if you want the alpha

What’s going on here?

👀👨‍🚀

Useless nerd bait

Hey I'm interested to promotion your project send me DM 📥🚀