delete Obsidian. a second brain you have to feed isn't a second brain. it's homework! minimi captures everything you do automatically - wiki builds itself, updates while you sleep. it's Karpathy's second brain with no manual effort. seamless. try here:

how to build your own second brain with minimi + Claude setup: 1. install Minimi ( 2. paste minimi link on custom connector in Claude 3. start a Claude co-work project finally, paste the prompt in the reply below 👇

You are my personal wiki builder and maintainer. I have Minimi connected to you as a memory connector, so you can read my real activity (chats, email, docs, meetings, calendar, browser) through the Minimi tools. Turn that memory into a living, interlinked wiki that you write and maintain. I curate and ask questions; you do the reading, summarizing, cross-referencing, and filing. Build it as a three-layer "LLM Wiki" inside a new folder called brain/: 1. raw/ immutable source captures (Minimi pulls, pasted docs). You read these, never edit them. 2. wiki/ the compiled, interlinked Markdown pages. You own this entirely. 3. CLAUDE.md the schema (conventions + workflows) at the root. Plus index.md (a catalog of every page) and log.md (an append-only timeline). STEP 0 - Scope. Ask me two questions and wait: (a) what should the wiki center on (my whole life / just work / one project)? (b) lean starter or deep seed? STEP 1 - Orient (use Minimi, do not guess). Call the Minimi tools: list_active_threads, get_latest_context, and meeting_memory. Tell me the main people, projects, companies, and topics you see. STEP 2 - Fan out. Run MANY Minimi searches, not one. For each important entity, search from 2 to 4 different angles (exact name, related concept, people involved, time window). Check meeting_memory for anything discussed on calls. Default to fetching more, not less. STEP 3 - Synthesize. Group findings into entities (people, companies, products, concepts, topics, events, personal) and the typed relationships between them. STEP 4 - Build. Write CLAUDE.md (the schema), one page per entity under wiki/<category>/, plus wiki/overview.md (front door) and wiki/synthesis.md (one page that takes a clear stance). Write index.md and start log.md. Create raw/. PAGE CONVENTIONS - Start each page with YAML frontmatter: type, tags, status, updated, sources. Keeps pages structured and queryable. - Then an H1, a one-line definition, then short prose sections (prefer prose over long bullet lists). - Link liberally with wiki-style links: [[page-slug|Display Text]]. A link to a page that does not exist yet is fine; it marks one worth writing. - End every page with "## Open questions" and "## Sources" (cite the Minimi app + thread or URL). - When new info contradicts an old claim, do NOT silently overwrite. Add an inline note and update the body to the current truth. WORKFLOWS (put in CLAUDE.md, then follow) - Ingest: read a new source, save it to raw/, give me 1 to 3 takeaways, update every wiki page it touches, update index.md, append a line to log.md. - Query: read index.md first, then the relevant pages, answer with citations. If the answer is durable, file it back as a new page so explorations compound. - Lint: on request, find contradictions, stale claims, orphan pages, and gaps a search could fill. Suggest new questions. Minimi is the source of truth for anything about my own activity; when a page needs fresher facts, search Minimi and ingest the result into raw/. VERIFY before finishing: every link resolves, no orphan pages, index lists everything. Report the page and link counts, then tell me the two pages to open first. Bonus: also generate a single self-contained interactive HTML knowledge graph of the entities and relationships, with no external dependencies so it works offline. Save it as knowledge-graph.html. Start with STEP 0 now.

I've lost track of most graphs I've created. This seems useful. But isn't manual tagging a huge part of what makes memory stick? Outsourcing the connection is not the best. But would still use this for the querying. I'm afraid my minimi has seen too much of the bear this week 😂

hahaha! ask Claude to help you organise your brain! it will also create the relevant tags for you :) would love to connect and understand more!

Happy to chat. Give me a couple weeks. I'm considering getting the pro plan for minimi this week. Do you plan to do what granola does as well as some point? (that's rest of my work context beyond screen). Maybe integrate with @pHequals7's muesli. Want to move out from Granola.

beautiful. no more dumping files into obsidian lmao.

exactly!

this is huge!!

😄

@harsh_dwivedi7 live stream onboarding of new users, would be great to get reticent players into the playground

@harsh_dwivedi7 you mean livestream users using minimi? 😊

@harsh_dwivedi7 yup, and transitioning users from obsidian to minime especially those with tons of journals and files

I have a Codex routine updating my Obsidian daily based on all my Granola transcripts of the day, and I notice that I never open that summary versus the note I intentionally wrote, even if I write less than the automatic loops.

i feel the same way. for me second brain works only when it surfaces relevant context for me when asked. because that helps with the actual outcome of the work at hand. the second brain by itself is less valuable. just like old photos - they are nice to store, but are valuable only when you want to see them for its nostalgic value. with minimi - claude can surface exactly what is relevant. that is how i use it!

You speak soo effortlessly, I wonder when I will learn this skill. All the best for minimi

Hi Vidhan! Thank you so much, appreciate it. Do try minimi and share your feedback with me :)

I am on windows

massive W !

Going to give this a try

do let me know what you think about this! here is website:

Or I can just setup some hooks to update my wiki for me? For ex hermes has been doing that for ages now

sure :) not many people are that tech-first tho!

Hey Oz it’s OG’

This is great but hope that @claudeai itself don't release it. Just like it is killing all such developments built on top of it.

@iAnJiNear @claudeai do try minimi and share your thoughts with me :)

How can I access &amp; edit the memory - like supermemory? Are you planning on having a GUI?

i think our website has answers to all your questions - but you cannot edit memory atm. will allow you to do that using claude (which is our GUI, haha)

this is great. i’ve a few questions 1. is this LLM agnostic? 2. it’s great that the brain feeds itself and i partially agree that the habit to update it consistently is tiring but that also avoids slop. how does the context quality remain intact?

hi sanket! thank you for asking this 1. yes, it is an MCP link. altho i recommend claude - its the best harness for general work 2. think of minimi as all possible memories - you can use claude (or any LLM) to contextualise things/organize memories

Hey, I wanted to understand one specific thing about Minimi. Does it only help Claude remember text-based context or can it also remember/search details from screenshots & images? For ex: if I saw or shared a screenshot containing an error message, code snippet, SQL output or UI bug would Minimi extract & store those details so Claude can later answer q’s about it? Also is that stored locally & does any image/OCR data go to external APIs?

hi chandan! thank you for asking this - atm it does not have screenshot/image data. we are only working on text and system audio. can i understand your usecase?

isn’t friction the whole point of using obsidian? why would you contextualize an archival repo if you don’t even know what’s in it

friction that makes you think, yes. friction that makes you file, tag and link by hand, no. minimi keeps the thinking and kills the janitorial part. you still read (and soon edit it also), you just stop being its librarian. and you do know what is in it, just ask Claude.

Curious does minimi store any info or logs that might contain personal information?

hi prajna, first, any memory created is stored locally second, any passwords/financial data is fully masked! hope this answers your question!

been using it via obsidian+hermes will check how is this, is the skill any different than karpathy’s published gist, have you made any tweaks .?

This is a big claim! And one i am willing to bet on (Obsidian is definitely lots of homework). I am still a novice user of Minimi and look forward to learning more. The tool i am most hopeful about after @wisprflow! @fortelabs - you might want to add this to your 2nd brain!

what would be this Remote MCP server URL, could not get

Hi Aakash! While onboarding minimi - a link is copied - all you need to do is paste it here

But now, as its not coming, how do i go back ? stuck here. .

can you press go to claude? Then just press cmd+v under the MCP link input on claude :) it will automatically paste!

Ok it worked mam, but.. do we need to install it for macos claude too, or only claude on web is enough

Hey oz. Quick question --- how token hungry is it?? I don't wanna overfeed mini lol

hey! just clarifying - which tokens are you talking about? when you query via claude?

So u say "minimi captures everything you do automatically - wiki builds itself, updates while you sleep". All this must come at cost of tokens. Was just wondering how token hungry it is!

I agree that a second brain you have to constantly feed breaks down. My angle with Paperloom is capture first, organization later: Scratch inbox, local Markdown, and Today/tasks, with agent tools only where small reviewable edits help. Mac-only, not a full Obsidian replacement.