Obsidian Memory Vault for AI Agents

This vault can be used as a long-term external memory system for any AI agent. Inspired by Andrej Karpathy's LLM Wiki: https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f

Purpose

• Preserve durable knowledge beyond the active model context window
• Store distilled, reusable knowledge instead of raw chat dumps
• Keep project knowledge, user preferences, procedures, decisions, and references searchable and linked

Start here

• [[Home]] — human-friendly entry point
• [[index]] — catalog of durable knowledge
• [[Memory Architecture]] — memory model and design principles
• [[log]] — append-only operational history
• [[AGENTS]] — operating contract for how an AI agent should use this vault

Main structure

• inbox/ — temporary capture and staging
• sessions/ — session history and episodic memory
• projects/ — canonical project notes
• concepts/ — stable ideas and definitions
• entities/ — people, orgs, tools, systems, products
• references/ — source notes, summaries, documents, transcripts
• procedures/ — reusable workflows and playbooks
• decisions/ — important decisions and rationale
• analyses/ — syntheses, comparisons, and question-driven writeups
• raw/ — immutable source material and attachments
• templates/ — reusable note templates

How an AI agent uses this vault

1. Capture incoming information
2. Classify it by memory type
3. Distill the durable signal
4. Update an existing canonical note when possible
5. Create a new note only when needed
6. Add wikilinks to related notes
7. Update indexes when a durable note is added
8. Append important maintenance actions to [[log]]

Management policy

This vault currently uses a hybrid management model:

• the agent may autonomously maintain routine notes, links, indexes, and memory hygiene
• the agent should ask before major restructures, bulk renames, bulk deletions, or schema changes

See also:

• [[Vault Management Policy]]
• [[Memory Ingestion Workflow]]

Vault activation prompts (What you are looking for)

Clone this repo 'long-term-agent-memory' and copy the folder 'long-term-agent-memory' to your project root, then use one of these ready-made prompts when you want an AI agent to adopt this vault as the active external memory system for your project.

Normal activation prompt

Use when starting a new session or when you want standard vault-backed work without a full audit.

Use the 'long-term-agent-memory' folder’s Obsidian vault as canonical external memory. Read the root docs first and follow the documented hybrid memory-management workflow for this session.

Deeper index-loading prompt

Use when you want broader context before starting work, but not a full vault-wide audit.

Use the 'long-term-agent-memory' folder’s Obsidian vault as canonical external memory. Read the root docs and all major index notes before continuing, then use the documented hybrid memory workflow for this session.

Full vault audit prompt

Use after a long gap, before cleanup, after major changes, or when you want the agent to refresh its understanding of the whole vault.

Use the 'long-term-agent-memory' folder’s Obsidian vault as canonical external memory. Audit the whole vault, refresh your map of the existing notes and structure, then continue using the documented hybrid memory workflow.

Recommendation:

• default to the normal activation prompt
• use the deeper index-loading prompt when you want more context up front
• use the full vault audit prompt only when broader reorientation is actually useful

See also:

• [[Vault Activation Prompts]]

Notes

• Do not store secrets in plaintext
• Prefer canonical notes over duplicates
• Prefer durable summaries over raw transcripts
• Use Obsidian wikilinks to keep the graph connected