Building a Knowledge Graph in Obsidian — Agent Links What I Can't Remember

The Problem

I talk to 31 agents across 8 teams. Every day, hundreds of decisions get made, people get mentioned, projects evolve. My brain retains maybe 10% of it. The other 90% dissolves.

Two months later: "What did we decide about the MiCA framework?" I know we discussed it. I know it was important. I cannot, for the life of me, remember the conclusion.

Manual note-taking doesn't work at my scale. I've tried. You open a note, type three bullets, get pulled into the next conversation, and never touch that note again. Two days later it's archaeologically buried.

The information exists — scattered across thousands of agent conversations. But without structure, without links, without a graph, it's noise. I needed a system that captures without my involvement and connects without my memory.

The Solution

Alfrawd captures every significant decision, entity mention, and project reference into an Obsidian vault. Automatically. With backlinks.

Not every message — that would be noise. Significant decisions, new relationships between entities, project state changes. The agent applies judgment about what's worth capturing, then files it with proper [[wiki-links]] so the knowledge graph builds itself.

The Process

The Obsidian vault lives on the Mac node, synced via iCloud. The agent writes markdown files directly:

# Agent creates/updates a note on Mac node
cat >> "/Users/bilal/Obsidian/PyratzLabs/People/Christophe POSTY.md" << 'EOF'

## 2026-03-10 — Stealth AM Discussion
- Discussed fund structure with [[Warren]]
- Connected to [[Stealth AM]] launch timeline
- Regulatory dependency: [[RCCI-RCSI]] certification
- Next step: Review with [[EuroNext]] filing team
EOF

Every entity gets its own note. People, projects, companies, regulatory frameworks. The backlinks create the graph:

[[Christophe POSTY]] → [[Stealth AM]] → [[Warren]] → [[RCCI-RCSI]]

When I ask "show me all notes connected to EuroNext filing," the agent queries the vault:

# Search for backlinks to a specific note
grep -rl "\\[\\[EuroNext\\]\\]" /Users/bilal/Obsidian/PyratzLabs/ \
  | while read f; do
      echo "---"
      echo "File: $(basename "$f" .md)"
      head -20 "$f"
    done

The visual graph in Obsidian shows the connections — nodes and edges mapping the entire decision landscape.

For deep search across months of context:

Me: what do I know about MiCA?

Alfrawd: Found 14 notes mentioning [[MiCA]] across 6 months:
- 2025-09-15: Initial MiCA impact assessment (Warren)
- 2025-10-02: MiCA compliance checklist drafted
- 2025-11-18: Decision to delay token launch until MiCA clarity
- 2026-01-09: MiCA framework update from EU Parliament
- ... [10 more entries]

Six months of conversations, surfaced in 3 seconds. With context. With the reasoning behind each decision.

The Results

The backlink density is what makes this powerful. After 6 months, the graph has enough edges that unexpected connections surface. "Wait, Christophe is connected to both Stealth AM and that regulatory issue we flagged in November?" That's the kind of insight that only emerges from a linked graph — never from linear notes.

Try It Yourself

1. Create an Obsidian vault with folders: People/, Projects/, Decisions/, Companies/
2. Configure your agent to write markdown with [[backlinks]] after significant conversations
3. Use daily notes for raw capture, entity notes for structured knowledge
4. Query the vault via grep, Obsidian search, or the agent's own search capability

The vault compounds. Month one feels sparse. Month six feels like a superpower.

Your memory is unreliable. Your file system isn't. Outsource accordingly.