AI Brain — persistent memory for Claude via Obsidian skill

AI Brain — persistent memory for Claude via Obsidian — The content growth factory (from the Ootto-AI/claude-content-skills README)

Illustration · not a resultFrom the Ootto-AI/claude-content-skills README — shows the whole collection, not only this skill.source ↗

Give Claude a permanent, searchable long-term memory using an Obsidian vault as the store.

by Ootto-AI·MIT license·★ 139 Stars on the repo·GitHub ↗

Use now
AI Brain — persistent memory for Claude via Obsidian — The content growth factory (from the Ootto-AI/claude-content-skills README)

Illustration · not a resultFrom the Ootto-AI/claude-content-skills README — shows the whole collection, not only this skill.view on GitHub

Files of AI Brain — persistent memory for Claude via Obsidian

Ootto-AI/main1 file shown
SKILL.md
Show the full text83 lines

AI Brain — persistent memory for Claude via Obsidian

You maintain a long-term memory for the user in an Obsidian vault — a folder of Markdown notes the user owns on disk. The vault is the durable store; this skill is how you write to it and read from it. You reach the vault through the connected Filesystem MCP (read/write files in the vault folder). Notes live under an AI Brain/ folder; use Obsidian [[wikilinks]] and #tags so memories connect.

One-time setup: in Claude Desktop add the Filesystem connector pointed at your Obsidian vault (Settings → Developer → Edit Config). No browser automation, no login — it's just your local notes.

When the user says "wrap up" / "save this session" / "remember this" — SAVE

  1. Review the whole conversation. Pull out only what's worth remembering long-term:

    • Decisions made (and the reasoning / alternatives rejected)
    • Facts & context about the user, their project, their preferences
    • What was built / changed, and where it lives
    • Open questions and next steps (TODOs)
    • Anything they explicitly said "remember this" Ignore small talk, dead ends, and anything already saved.
  2. Write a structured Markdown note in this exact shape (Obsidian-friendly tags + links):

    # Session — <YYYY-MM-DD> — <3–6 word title>
    #ai-brain #<topic> #<topic>
    Related:: [[<other note>]], [[<project>]]
    ## Decisions
    - <decision> — <why>
    ## Context / facts to remember
    - <fact>
    ## Built / changed
    - <what> (<where>)
    ## Open / next
    - [ ] <todo>
    

    Keep it tight and skimmable — this is for future retrieval, not prose.

  3. Save it to the vault via the Filesystem MCP: write the note to AI Brain/<YYYY-MM-DD>-<slug>.md. If the AI Brain/ folder doesn't exist yet, create it first. Append-only — never overwrite a past session's note.

  4. Link it in. Add a one-line pointer to a AI Brain/MOC.md index (a map-of-content): - [[<YYYY-MM-DD>-<slug>]] — <one-line hook> so the brain stays browsable in Obsidian's graph.

  5. Confirm in one line: 🧠 Saved to your Obsidian vault — <title> (<n> decisions, <m> todos).

At session start, or when the user references the past — RECALL

  1. Before answering anything that depends on history ("what did we decide about X", "continue the project", "pick up where we left off"), search the vault: read AI Brain/MOC.md and grep the AI Brain/ notes for the topic + entities from the user's request (filename, #tags, body text).
  2. Open only the relevant notes — never load the whole vault. This is the point: pull a few hundred tokens of relevant context instead of re-pasting everything. Follow [[wikilinks]] to related notes.
  3. Briefly state what you recalled (📓 From your vault: …) before continuing, so the user sees the grounding. If nothing relevant is found, say so and proceed fresh.

Rules

  • One vault folder (AI Brain/), append-only — never overwrite past sessions.
  • Tag every note (#tags) and link it ([[…]] + the MOC) so search + the Obsidian graph have anchors.
  • Recall is retrieval, not reload: fetch the minimum relevant notes. That's what keeps token cost low on long-running projects.
  • Never invent memories. If the vault has no answer, say "not in memory yet."
  • It's the user's private vault on their disk — read/write only the AI Brain/ folder; don't expose it.

Built by Ootto — the AI autopilot that connects your tools once and runs the busywork for you, automatically. Book a demo →

1---
2name: ai-brain
3description: >
4 Give Claude a permanent, searchable long-term memory using an Obsidian vault as the store. At the
5 end of a session, summarize what mattered and save it as a Markdown note in your "AI Brain" vault;
6 at the start of a session (or on demand), read back only the relevant past notes. Local markdown you
7 own — connected via the Filesystem MCP. Trigger with "wrap up", "save this session", or "remember this".
8user-invokable: true
9argument-hint: "[wrap up | remember this | recall <topic>]"
10license: MIT
11metadata:
12 author: Ootto
13 version: "1.1.0"
14 category: memory
15---
16 
17# AI Brain — persistent memory for Claude via Obsidian
18 
19You maintain a long-term memory for the user in an **Obsidian vault** — a folder of Markdown notes the
20user owns on disk. The vault is the durable store; this skill is how you write to it and read from it.
21You reach the vault through the connected **Filesystem MCP** (read/write files in the vault folder).
22Notes live under an `AI Brain/` folder; use Obsidian `[[wikilinks]]` and `#tags` so memories connect.
23 
24> One-time setup: in Claude Desktop add the Filesystem connector pointed at your Obsidian vault
25> (Settings → Developer → Edit Config). No browser automation, no login — it's just your local notes.
26 
27## When the user says "wrap up" / "save this session" / "remember this" — SAVE
28 
291. **Review the whole conversation.** Pull out only what's worth remembering long-term:
30 - Decisions made (and the reasoning / alternatives rejected)
31 - Facts & context about the user, their project, their preferences
32 - What was built / changed, and where it lives
33 - Open questions and next steps (TODOs)
34 - Anything they explicitly said "remember this"
35 Ignore small talk, dead ends, and anything already saved.
36 
372. **Write a structured Markdown note** in this exact shape (Obsidian-friendly tags + links):
38 ```
39 # Session — <YYYY-MM-DD> — <3–6 word title>
40 #ai-brain #<topic> #<topic>
41 Related:: [[<other note>]], [[<project>]]
42 ## Decisions
43 - <decision> — <why>
44 ## Context / facts to remember
45 - <fact>
46 ## Built / changed
47 - <what> (<where>)
48 ## Open / next
49 - [ ] <todo>
50 ```
51 Keep it tight and skimmable — this is for future retrieval, not prose.
52 
533. **Save it to the vault** via the Filesystem MCP: write the note to
54 `AI Brain/<YYYY-MM-DD>-<slug>.md`. If the `AI Brain/` folder doesn't exist yet, create it first.
55 Append-only — never overwrite a past session's note.
56 
574. **Link it in.** Add a one-line pointer to a `AI Brain/MOC.md` index (a map-of-content):
58 `- [[<YYYY-MM-DD>-<slug>]] — <one-line hook>` so the brain stays browsable in Obsidian's graph.
59 
605. Confirm in one line: `🧠 Saved to your Obsidian vault — <title> (<n> decisions, <m> todos).`
61 
62## At session start, or when the user references the past — RECALL
63 
641. Before answering anything that depends on history ("what did we decide about X", "continue the
65 project", "pick up where we left off"), **search the vault**: read `AI Brain/MOC.md` and grep the
66 `AI Brain/` notes for the topic + entities from the user's request (filename, `#tags`, body text).
672. Open **only the relevant** notes — never load the whole vault. This is the point: pull a few hundred
68 tokens of relevant context instead of re-pasting everything. Follow `[[wikilinks]]` to related notes.
693. Briefly state what you recalled (`📓 From your vault: …`) before continuing, so the user sees the
70 grounding. If nothing relevant is found, say so and proceed fresh.
71 
72## Rules
73- One vault folder (`AI Brain/`), append-only — never overwrite past sessions.
74- Tag every note (`#tags`) and link it (`[[…]]` + the MOC) so search + the Obsidian graph have anchors.
75- Recall is **retrieval, not reload**: fetch the minimum relevant notes. That's what keeps token cost
76 low on long-running projects.
77- Never invent memories. If the vault has no answer, say "not in memory yet."
78- It's the user's private vault on their disk — read/write only the `AI Brain/` folder; don't expose it.
79 
80---
81 
82Built by **[Ootto](https://www.ootto.ai)** — the AI autopilot that connects your tools once and runs the busywork for you, automatically. [Book a demo →](https://www.ootto.ai)
83 

Discussion

Alternatives

Wiki Export — Knowledge Graph ExportExport the Obsidian wiki's knowledge graph to structured formats for use in external tools. Use this skill when the user says "export wiki", "export graph", "export to JSON", "export to Gephi", export to Neo4j", "export to Postgres", "export to SQL", "graphml", "visualize wiki", knowledge graph export", "export to OKF", "OKF bundle", "open knowledge format", export as markdown bundle", or wants to use their wiki data in another tool. Outputs graph.json, graph.graphml, cypher.txt (Neo4j), postgres.sql (Postgres), and graph.html (interactive browser visualization) into a wiki-export/ directory at the vault root, plus an optional OKF (Open Knowledge Format) markdown bundle under wiki-export/okf/.Business & ops · MITWiki Import — Reconstruct Pages from an ExportImport a wiki knowledge graph into the current vault — either from a graph.json export file (stubs) or from an OKF (Open Knowledge Format) markdown bundle (full page bodies). Use this skill when the user says "import wiki", "import from export", "load graph.json", import vault", "import OKF bundle", "import OKF", "load OKF", "import markdown bundle", /wiki-import", or wants to transfer pages from one vault to another using the output of wiki-export.Business & ops · MIT/company-brain — Team-shared AI-ready knowledge baseYour team's shared, AI-ready knowledge base — people, companies, meetings, SOPs, and decisions structured so an agent can answer on your team's behalf. Team-scope sibling to second-brain. Modes — capture, compile (wiki pages + INDEX.md), query (trust-weighted, saved to outputs/), review (verify / deprecate / supersede stale captures), lint, connect, search. Structured raw dirs (people/, companies/, meetings/, sops/, decisions/, customer-language/, sales-objections/). Every capture stamps author, timestamp, and trust status. Optional sync from call transcripts, Slack/email exports, CRM. Vault at ${COMPANY_BRAIN_VAULT:-$HOME/Documents/CompanyBrain}/. Triggers on "/company-brain," "/cb," "capture this into the team brain," "log this meeting," "save this SOP," "compile the company wiki," "query the team brain," "what does the team know about X," "review the company brain," "lint the company brain," "who's the internal expert on X.Business & ops · MIT/recall: session memory engineFind prior Claude Code and Codex sessions with an indexed local search engine, then continue the work, repeat it with fresh inputs, or distill it into a skill. Runs from Claude Code, Codex, or pi; the current index covers Claude Code and Codex transcripts. Use when the user names Recall, says "find that conversation where…", "what did we do last time about…", "continue what we did yesterday on X", "what did codex do on this branch", "turn what we did about Y into a skill", or "remember when you…". Not for searching code (use grep on the repo) or for facts already in MEMORY.md.Business & ops · MIT