Back to all resources
RESOURCE LIVE

Give every agent the same memory.

GBrain is Garry Tan's open-source memory layer. Markdown notes that wire themselves into a graph any agent can read, so Claude Code, Cursor and everything else stop starting from zero. Two minutes to install. Your existing Obsidian vault comes with you.

Raj Singh By Raj Singh · August 2026 · 7 min read
SHARED MEMORY, LIVEOne graph, every agent reading it
gbrain · search 'meridian'
7 nodes · 12ms
Client note — Meridian prefers Loom over calls. Dana decides.
Linked — Proposal thread, two invoices, last week's call notes.
Read by — Claude Code, Cursor and the morning brief, same graph.
Captured once. Available everywhere.Self-wiring · MCP served
THE TELL

Every agent you own starts the day knowing nothing.

  • You run Claude Code, Cursor and two others. None of them share a single fact about your business.
  • Every morning you re-explain the company to whichever agent you opened first. That briefing is the same briefing, every time.
  • Obsidian is excellent right up to the point where your agents cannot read it, which is where you actually needed it.
  • The notes you take and the context your agents need have been two separate piles this whole time.
WHAT'S INSIDE

Install, capture, import, serve.

  • 01
    Two-minute install
    bun, gbrain init and the ZeroEntropy embedder, with the one flag that decides how the graph wires itself.
  • 02
    Capture and think
    The daily loop that keeps the graph current without turning note-taking into a second job.
  • 03
    Your Obsidian vault
    Importing what you already have without re-tagging a single note, and what GBrain infers from the existing links.
  • 04
    Served over MCP
    One command turns the graph into an MCP server, and every agent you run gets the same memory.
  • 05
    When it drifts
    The doctor script that finds orphaned nodes and broken links before your agents start citing things that moved.
BUILT FOR

You, if you brief the same facts every single day.

If more than one agent touches your work, the re-briefing tax is already the most repetitive thing in your week. This removes it in an afternoon.

Why now Self-wiring memory only became practical once embedding costs fell far enough to re-index a whole vault continuously. Manual knowledge graphs never survived contact with a real week.

The change is that context stops being something you carry between tools. You say a thing once, and every agent you own already knows it tomorrow.

A woman sitting at a desk with a laptop
Photo: Mykyta Kravčenko / Unsplash
UNLOCKS WITH YOUR EMAIL
01install.md1.8 KB
02capture-think.md2.4 KB
03mcp-config.json0.3 KB
04obsidian-import.md1.6 KB
05gbrain-doctor.sh0.5 KB
Get the resource

Drop your email, get the full tool.

One email. Instant access. No sequence, no upsell, no list-trap. I'll only ping you if I ship something that fits.

One email. No follow-up unless I ship something that fits.

The toolkit: turn your Markdown notes into a knowledge graph every agent on your machine can read. GBrain is Garry Tan's open-source memory layer — local-first, embedding-agnostic, and MCP-native so Claude Code, Cursor, and OpenClaw all share the same brain.

  • install.md is the 2-minute setup. `bun install`, `gbrain init`, drop in a ZeroEntropy or local-Ollama embedder, you are running. Zero config beyond the embedder key.
  • capture-think.md is the day-to-day workflow. `gbrain capture` for new notes, `gbrain search` for retrieval, `gbrain think` for the multi-hop synthesis that actually feels agentic. Two commands, every day.
  • obsidian-import.md moves an existing Obsidian vault into GBrain without re-tagging. The wikilinks become graph edges automatically — your old vault is suddenly queryable by every agent.
  • mcp-config.json is the one file that wires GBrain into every agent. Point Claude Code at it (or Cursor, OpenClaw, Hermes) and ask any agent a question — it gets the brain for free.
  • gbrain-doctor.sh is the diagnostic. Stale embeddings, broken links, missing chunks — run it weekly and the brain stays clean. No more "why does Claude not know about X."
  • macOS, Linux, or WSL2. Windows native needs a Docker container (a one-line setup is in install.md)
  • Bun 1.1+ runtime
  • An embedder choice: ZeroEntropy (fastest, paid), Ollama with nomic-embed (free, local), or OpenAI text-embedding-3 (cheap, fast)
  • 5 minutes from `git clone` to first search. The vault scale is whatever your disk holds

↑ Unlocks the moment you submit the form above.

RS
WRITTEN BY

Raj Singh

I'm Raj Singh, founder of DebtOps.ai and a Claude certified partner. Over the last five years I've managed $25M in ads, built AI solutions for Fortune 100 companies, and installed AI operating systems and AI brains inside small businesses. Everything here is something I've shipped, debugged, or watched a client ship. My goal with AgileSingh is to make you irreplaceable in this AI age.

$25M
Ads managed · 5 yrs
F100
Enterprise AI builds
Claude
Certified partner
18
Guides shipped
More guides

Now point a swarm at it.

Shared memory matters most when there is more than one agent reading it. The agent swarm guide is where that compounds, and the MCP guide covers the serving layer. Browse the library, or pin a slot and I will set yours up.