Ari’s Space*   est. 2026
Skip to content
ari*s space[ a. ]
← Librarymemory / 2026-08-06

I built a public version of my brain for the website

Ari's Brain is a Cortex-shaped public consult over my teaching library so people (and their agents) can learn AI better from the site as a knowledge bank.

10 parts / Hover to explore. Click to read.

Ari's Brain is a Cortex-shaped public consult over my teaching library so people (and their agents) can learn AI better from the site as a knowledge bank.

Personal notes -> own Cortex
                  remember/recall

Published essays -> teaching index
                         | recall
                    source essay

No personal writes cross over.
The planned member consult reads published teaching through a separate server and permission boundary. Live attachment remains coming soon until the public credential path exists; feedback never writes to the teaching projection.
Context Pack (for agents)

Cortex on my own desk helps a session begin with useful memory. The files keep the source material; the brain supplies connections the next model can use.

That personal system contains things which should stay on my machine. The teaching I publish on Ari's Space is a different body of work, already intended for other people to read.

I wanted an agent to be able to consult that teaching instead of guessing at a half-remembered version of it. That's the job of Ari's Brain: a public-teaching counterpart with the same general shape as Cortex and a different boundary around what it can access.

This article describes the member-consult design. Live attachment remains a shipping-soon feature until a public member-credential path is available. The free essays can already be read without that tool.

I put a Cortex-shaped brain on the public site so people can learn from the work I already taught.
PERSONAL window with folder, dual arrow to CONSULT window with open book, then steps to a document page
Personal stays local. Teaching gets a consult. The site becomes a knowledge bank you can stand behind.

Make the teaching usable inside the work

The library covers memory layers, context packs, seats, labour and the craft of running these systems. It's useful in a browser. I also want it to be useful when your agent is helping you build something.

That means giving the agent a way to find relevant teaching and take you back to the source. Otherwise the careful explanation can sit in a tab while the conversation invents an approximate version from elsewhere.

The aim is to help people learn AI better. Membership adds a tool for consulting the work. It doesn't put the essays themselves behind a gate.

What the consult contains

PartRole in the design
Published libraryThe essays and methods I've chosen to make public.
Ari's BrainA remote member consult, with a planned soft ARCHITECT+ access level when live.
ToolsThe familiar Cortex family: recall, regions, modes, lineage and feedback.
Publishing boundaryNo public remember. I publish the material; the index rebuilds from it. Feedback reports problems separately.

Think of the site as a knowledge bank your tools can consult. It still needs a clear source, a way to retrieve useful material and a limit on who can change that material.

Keep the two stores separate

Personal Cortex remembers your work. The teaching consult retrieves published material. An agent can attach to both, but they must remain separate servers with separate permissions.

Dual attach: personal Cortex and Ari's Brain as two MCP serversOne agent window in the centre. Two System 7 server windows sit left and right. Left is personal Cortex, local, with remember and recall. Right is Ari's Brain, remote teaching, with recall and feedback only. Same tool family names, different server IDs, no shared write path into the teaching index.PERSONAL CORTEXserver: ari-os-cortexwhere: your machineyour notes · your vaultbrain.rememberbrain.recallmodes · regions · dreamwrites stay localAGENTtwoattachesone seatARI'S BRAINserver: aris-brainwhere: remote consultpublished teachingbrain.recallbrain.lineagebrain.feedbackno public remembersame tool family · different server · different trust planepersonal writes never land in the teaching index
QuestionPersonal CortexAri's Brain
What is it for?Remembering you and your work.Consulting the teaching library.
Where does it live in this design?Your machine.The remote member consult.
What is authoritative?Your memory files.The published library.
Can the agent add knowledge?Through the personal memory write path, when authorised.No public write into the teaching index. Feedback has a separate path.
Who benefits?The person using the local system.People learning from the published corpus.

The index on either side is derived from its source files. It should be rebuildable without losing the canonical knowledge. That is different from saying every operational log or feedback record is disposable.

Similar tool names help an agent use both systems. Distinct server IDs keep it clear which system a call is reaching. Personal writes must never land in the teaching index.

Retrieve the material, then follow it home

A consult returns ranked units from the library. It isn't meant to replace the essay with a freshly invented explanation.

The read path is: call brain.recall, retrieve relevant material from the library projection, inspect the returned snippets and regions, then use brain.lineage to reach the source essay.

The projection is a derived index of published material. The design records its rebuild state and returns source hashes so the material has an identifiable origin. Member credentials, rate limits and response budgets apply to the consult surface.

Read path from agent through MCP to projection and essay lineageVertical System 7 pipeline. YOUR AGENT calls brain.recall on the aris-brain MCP server. The server hits the library projection. Ranked units return with region and snippet. brain.lineage walks a unit back to the public essay URL. Feedback is a side channel that never writes the projection.YOUR AGENTstuck on an AI systems questionbrain.recall(query)MCP · aris-brainmember credential · rate limitedquery the projectionLIBRARY PROJECTIONpublished essays · rebuild stampedfiles own truth · index is derivedranked units + source_hashUNITS TO AGENTsnippet · region · mode-rankedbudget capped · no invented bodybrain.lineage(unit_id)ESSAY URLslug · public documents pathwalk home. always.FEEDBACKsidechannelnot index

If an agent quotes a method, you should be able to open the page that supports it. A plausible answer with no way back to the source isn't enough.

Feedback belongs beside this path. It reports bad retrievals and gaps in the material; it does not write to the teaching projection.

Let modes change the retrieval stance

The personal system can have a larger set of modes. The public consult uses a smaller set appropriate to the published corpus.

Default aims for balanced everyday retrieval. Focus stays close to the task. Synthesis joins material across regions. Recall supports a quick lookup. Dyslexic keeps the returned material plain and easy to skim.

Modes as posture, not a dump of every dialFive small System 7 tiles in a row for default, focus, synthesis, recall, and dyslexic. Caption says same teaching corpus, different ranking stance.defaultbalancedeverydayfocustighton-tasksynthesisjoinacrossrecallfastlookupdyslexicplainskimmablesame teaching corpus · different ranking stancenot a dump of every personal dial

These are ways of approaching the same teaching. A mode doesn't change the source or grant permission to access a different person's private memory.

Don't make the private store the public product

My local Cortex contains private work, client material and unfinished notes. That is a different job from helping someone find a method I've already published.

Ari's Brain uses the second body of material. It doesn't require copying the first into a public service.

The separation is the useful part of the design: personal memory can stay local while published teaching becomes a second, gated consult. Someone who only wants to read the free library doesn't need either attachment.

Apply the same distinction to a business

Ask what knowledge you would want an agent to use when explaining your business to a client.

Start with material you deliberately stand behind: published support playbooks, product principles, explanations of how you work. That is a different corpus from an internal drive, old chat history or customer secrets.

The pattern is straightforward:

  1. Keep private operations private.
  2. Build a replaceable index over the material you've chosen to publish.
  3. Provide a surface that lets agents read it without giving everyone a write path into it.
  4. Return enough lineage for a claim to lead back to a page you own.

You don't need my implementation to make those decisions. You do need to know which source is authoritative and where the boundary sits.

Check availability before attaching

The shipping note matters: treat the live consult as coming soon until the member-credential path is public. This design essay isn't an attachment endpoint or a promise that the service is already available to your account.

When that path is ready, aris-brain can sit beside the local Cortex attachment. Keep the server IDs and permissions distinct. Membership is the tools-and-access layer; reading the teaching remains free.

Read the part that matches your next job

Start with Cortex for the personal memory philosophy and local pattern. If you're using ARI-OS, the local brain guide explains the operator surface.

Generative Memories explores a related concern: making provenance visible when a system learns from work that wasn't part of the original live project.

This essay applies that concern to a public consult. The agent needs to know what it retrieved, where it came from and which store it is allowed to change.

Attach two systems when you need both jobs: personal Cortex for your memory, Ari's Brain for the published teaching. Keep the sources separate and the path back to the essay easy to follow.

[ comments ]

guests welcome. members show first in the list

no comments yet. start the thread_

1 / 12