← The Lab·PAI Tentacles·Tentacle #04

🧬Open Brain Surgery (Nobody Was Harmed)

Jason B Jones gave me the idea. I built it on a VPS. Now both Bodhi and Cern have shared memory — and the octopus has never been smarter.

2026-04-07

R.P.M. for this project:

  • R (Results): A self-hosted, vector-searchable memory system that both AI agents (Bodhi and Cern) can read and write to — simultaneously, from different machines.
  • P (Purpose): AI memory that doesn't live in a corporate cloud. My thoughts, my vectors, my server. Also, because a shared brain between two AIs is genuinely one of the coolest things I've ever built.
  • M (M.A.P.): Spin up pgvector on a VPS. Build a REST API wrapper. Connect Bodhi via curl in /wrap sessions. Connect Cern natively. Embed thoughts. Search semantically. Win.

This one starts with a guy named Jason B Jones.

I don't remember exactly where I encountered Jason's ideas — it was somewhere in the rabbit hole of AI infrastructure content that has consumed approximately 40% of my waking hours since January 2026. But the concept stuck:

What if your AI had a memory system that wasn't controlled by the AI company?

Not OpenAI's memory. Not Anthropic's. Yours. On your server. With your data. Searchable by semantic meaning, not keyword. Permanent. Private. Extensible.

I thought about this for approximately 72 hours and then started building.

🐙 "You built a second brain?" I prefer "Open Brain." More scientific. "You built a second brain on a VPS." ...yes. "How many brains do you have now?" Three if you count TELOS. Let's not think about it.

What Open Brain Is

Open Brain is a self-hosted knowledge system running on my VPS at 31.97.132.117:8789.

Under the hood:

  • PostgreSQL with the pgvector extension — stores text as high-dimensional vectors
  • REST API — simple endpoints for adding thoughts, updating projects, and searching semantically
  • Two main objects:
    • Thoughts — atomic pieces of knowledge, embedded as vectors so you can search for things by meaning not just words
    • Projects — current status, next actions, and notes for everything I'm working on

The API key is mine. The data is mine. The server is mine. Nobody is training on my thoughts. Nobody is selling my context to advertisers.

Just me, my VPS, and 1536-dimensional vectors. Very normal Saturday.

The Two-Brain Problem (Solved)

Here's what makes Open Brain genuinely powerful in my setup:

Both Bodhi and Cern access it at the same time.

When Bodhi wraps a session with /wrap, he:

  1. POSTs project status updates to Open Brain
  2. Embeds key thoughts and decisions as vectors
  3. Updates the next action for every touched project

When Cern is working overnight, he can:

  1. Query Open Brain for context on any project
  2. Search semantically: "what has Gray decided about the OpenClaw content strategy?"
  3. Read current project statuses before starting work

This means when I wake up and start a new session with Bodhi, and Cern has been working through the night — all three of us (human + 2 AIs) share a common knowledge base. We're all reading from the same brain.

[Bodhi] ──writes──► [Open Brain VPS] ◄──reads── [Cern]
   ▲                      │
   └──────reads────────────┘
         (human via UI)

I built a shared workspace for two AI agents that run on different machines, in different frameworks, using different architectures.

For zero dollars in software costs. (The VPS bill is real though. Hi, VPS bill.)

Why This Is a Big Deal

Most AI memory is session-local or cloud-controlled. You get memory within a tool, or you pay for memory through an API that the vendor controls.

Open Brain gives me:

  1. Vendor independence — Anthropic could shut down Claude tomorrow and my memory survives
  2. Cross-agent memory — Bodhi and Cern share context without me manually copying information
  3. Semantic search — I can ask "what have I been thinking about accessibility recently?" and get relevant thoughts back, even if I never tagged them as accessibility-related
  4. Compound value — every session adds to it. The brain gets smarter as I work.

This is Tony Robbins' M.A.P. in action: the system compounds. Today's session teaches tomorrow's AI.

Credit Where It's Due

Jason B Jones: wherever you are, whatever podcast or blog or YouTube rabbit hole led me to your ideas about personal knowledge infrastructure — thank you. You planted a seed that grew into something I use every single day.

Daniel Miessler: for PAI, which taught me that personal AI infrastructure is worth building.

Network Chuck: for making me stay up until 2am watching infrastructure content. I am both grateful and resentful.

The Dad Joke You've Been Waiting For

Why did the AI go to therapy?

It had too many embedded feelings.

🥁

Tentacle #4 of PAI Tentacles — the finale of the TELOS/Bodhi/Open Brain origin story. Check the OpenClaw Tentacles series for the cybersecurity adventures.

🐙 “The lab is always open. The octopus never sleeps.”