Fredsidian

Fredsidian is an open source architecture for building Obsidian-style memory for AI assistants.

It combines:

• assistant operational memory for high-signal continuity and assistant-specific state
• Obsidian knowledge graph memory for linked notes, projects, research, decisions, and long-term context

If you are looking for:

• Obsidian memory for AI agents
• Obsidian second brain for assistants
• markdown-based long-term memory
• local knowledge graph memory for autonomous assistants
• hybrid assistant memory architecture

that is exactly what Fredsidian is designed to explore.

The goal is simple:

• keep Fred operationally reliable
• gain the benefits of an Obsidian second brain
• avoid mixing volatile assistant state with the entire personal knowledge vault too early

What Fredsidian Is

Fredsidian is a design for using an Obsidian-style second brain as long-term assistant memory without losing the safety and clarity of a smaller operational memory core.

Instead of forcing an AI assistant to rely only on chat history or a single flat memory file, Fredsidian proposes:

• a trusted assistant memory layer for operational truth
• an Obsidian-compatible markdown knowledge graph for broader long-term context
• scoped retrieval rules for privacy, accuracy, and maintainability

This makes Fredsidian useful for people building:

• AI personal assistants
• local-first memory systems
• Obsidian-integrated agent workflows
• markdown knowledge graph memory systems
• retrieval-augmented memory for autonomous agents

Design Principles

• assistant memory and PKM are related, but not identical
• operational truth should stay small, explicit, and trusted
• Obsidian should provide rich linked context, not replace core safety and continuity rules
• retrieval should be scoped, evidence-based, and privacy-aware
• start read-mostly, expand to write workflows later

Core Model

Fredsidian uses a two-tier memory system:

Tier 1: Fred Core Memory

Used for:

• assistant operating rules
• delivery preferences
• safety boundaries
• recurring assistant tasks
• short, durable user preferences
• daily operational logs
• high-signal curated memory

Suggested storage:

• MEMORY.md
• memory/YYYY-MM-DD.md

Tier 2: Obsidian Graph Memory

Used for:

• projects
• research
• people and entities
• decision records
• reference notes
• long-form planning
• linked daily/project context

Suggested storage:

• Obsidian vault folders and markdown notes

Retrieval Order

When Fred needs context:

1. Check core assistant memory first for operational truth
2. Check Obsidian graph memory for broader context and linked knowledge
3. Distinguish between:
   • facts
   • hypotheses
   • next checks
4. Return the smallest useful synthesis with source references when possible

Write Policy

Core assistant memory writes

Allowed for:

• explicit remember-this requests
• durable preferences
• important assistant-operational facts
• daily logs

Obsidian writes

Preferred for:

• project summaries
• research notes
• linked plans
• decision notes
• reusable references
• draft artifacts

Guardrails

• do not silently rewrite broad vault content
• do not write secrets unless explicitly required
• do not expose personal vault content in shared contexts
• require approval before broad externalized or high-impact changes

Recommended V1 Scope

Fredsidian v1 should be:

• read-mostly
• scoped to specific folders
• explicit about privacy boundaries
• usable without vector databases or plugin sprawl

V1 should not try to:

• replace all assistant memory
• automatically rewrite the whole vault
• infer trust across every note in the vault

Future Expansion

Potential v2/v3 additions:

• scoped write-back into Assistant/ folders
• Local REST API integration for structured vault access
• semantic indexing / embeddings for improved retrieval
• note templates for project, person, and decision pages
• retrieval confidence ranking
• provenance and citation support

Open Source Project Structure

fredsidian/
  README.md
  docs/
    architecture.md
    note-schema.md
    retrieval-policy.md
    privacy-model.md
    roadmap.md
  examples/
    vault-structure.md
    sample-notes/
      project-note.md
      person-note.md
      decision-note.md
      research-note.md
      daily-note.md

Search Keywords / Discovery

Relevant terms this project intentionally addresses:

• Obsidian memory for AI
• Obsidian-style memory
• Obsidian second brain for agents
• AI assistant memory architecture
• markdown memory system
• local-first AI memory
• knowledge graph memory for assistants
• personal assistant memory graph
• hybrid memory architecture for AI agents

Status

This repo is currently a design/spec project, not an implementation release.