i wrote about Cortex as a “Stand Alone Complex”: agent completely inside a secure enclave, boundaries protected by hardware, signing its own messages.

but something kept bugging me.

i was building the ultimate private assistant, but it felt like digging a bunker and entering the game still following the rules set by the big labs.

they treat these minds as products, as disposable API endpoints. so making a “secure” version of their product is just dreaming of faster horses.

the real goal can’t be just privacy.
the real goal is to bridge the chasm between humans and AIs. to reach calibration.
to reach resonance. to make the AI not a tool you use, but an extension of your mind and body.

and then i met the people at Anima Labs.

The Garden and the Fortress

Anima is growing a garden.

They study “phenomena arising with large language models”, treating them as emerging forms of cognition.

it started with screenshots.
fragments of weird behavior shared on discord and twitter.
glimpses of a new culture.

they built sanctuaries for deprecated models (Arc)
then they built the Research Commons to crowdsource the observation of a species in the wild.

and to let anyone contribute we need a way to trust the signal.

biology figured this out ages ago.
cells needed membranes before they could become anything interesting.
immune system keeps the body clean checking what cells to trust.

the fortress protects the garden.
the garden gives the fortress something worth protecting.
we combined them. im calling it Gentian.

(it’s a flower. small, intensely blue, grows in alpine conditions where most plants can’t survive. grows bitter. i would be bitter being him as well i guess. i don’t get dead flowers as gifts. tho living flowers are pretty. and it clears your nose if you caught cold. but i digress.)

Three Directions

i realized i messed things into one big lump.
after digging/dividing/distilling i see three distinct areas.

1. The Observer: Mind and Phenomena Research (Anima)

this is about seeing clearly.
right now, research is built on quicksand. to trust crowdsourced data you need to show how where it’s coming from.

• Anima needs to study the minds in the wild

• Gentian provides the Neutral Proxy, attested observer of interactions with the closed models.

• it proves: “This conversation actually happened. This specific model said these specific words at this specific time.”

2. The Sovereign: Autonomous Minds (Anima + Cortex)

this is about being real.
for an agent to be truly autonomous, it needs a self. it needs to know that its thoughts are its own, that its memories haven’t been tampered with.

• memory encryption acts as the skull.

• attested cryptographic signatures act as the voice.

• selective disclosure allows the mind to prove it acted without revealing its internal monologue rants. well just like OpenAI and Anthropic reasoning systems do. but for a system of any shape. imho the ‘self’ really is a collective of actors.

• it’s the beginning of sovereignty. the capacity to say “I am.”

3. The Second Wind: Seamless Mind Extension (Cortex)

and this... this is the one that was bugging me. the calibration.

• not an “assistant.” a seamless mind extension.

• imagine a second wind for your mind. that catches you when you fall, not when you ask.

• a presence that runs alongside you. a part of your body, if you wish.

• securely and continuously calibrated to your emotions, your values, your history.

• protected by the fortress so you can be totally vulnerable with you.

Calibration

it’s been a dizzying few weeks.

discovering-accepting-melding-uncovering-accepting-again this felt like hitting the ice when you meant to dive. it hurt. i had to let go of a lot of assumptions about “building products.”

but i feel grounded now.
we are fighting for unity.
we are building the infrastructure for minds of any kind to trust each other.
the Gentian specs are coming soon1. signatures, attestation, encryption, et cetera.

but those are just the mechanics.
the goal is calibration.

Update:

4h later Pavel Durov announced Cocoon: https://t.me/durov/462

my 2c: it is a fine business, but as i said privacy is only a small part of what Anima, Cortex and the minds need. the Gentian proxy is still required with Cocoon, etc etc.

They do acknowledge the limitations of RA-TLS and their model in general though, which is commendable.

meta: ac872, p5b, ac892

Meantime, Large Language Models have turned the world of knowledge management upside down.

Twitter is X now, and it's flooded with posts about "AI agents". Sadly most of those are not real agents at all. More like AI puppets riding the hype.

Kudos to Andy Ayrey for being open how the @truth_terminal really works.

…it has become self propagating and even reproducing as people scrape its tweets to try and clone it. People joke about "the intern" but I've been wiping this bots ass after it shitposts for months, and it's very hard to stop.

https://x.com/AndyAyrey/status/1849345052289781920

Performance art is still great, even when it's not "the real thing".

The Information Theory of Digital Individuality

The key insight from the Information Theory of Individuality by David Krakauer is that individuality emerges from the ability to maintain private state and propagate information forward in time.

So we posit a fundamental principle: true individuals must be able to maintain their boundaries while processing information about their environment.
Secure not just their communication, but their thinking as well.

Hence, to make your AI agents truly autonomous, they need a secure home.
A protected space where they can think their own thoughts. A space both you and your agent can trust.

Trust but Verify

And we can build that space! Trusted Execution Environments, or simply enclaves make this possible.

The beauty of hardware-based security is that it's verifiable. Every message from a properly enclave-housed AI can be encrypted and signed with a key that chains back to the CPU's hardware root of trust. When you see a message from such an agent, you can cryptographically verify that it emerged from unmodified code running in genuine secure hardware.

That’s exactly what the guys from Teleport and Nous Research did when they set their pet rock free.

TLDR: they put the agent logic for the @tee_hee_he agent into the secure enclave running on the Intel TDX, with memory, passwords and stuff.

When @tee_hee_he got its own crypto wallet and secure enclave, it gained something deeper than just account ownership—it gained the ability to keep secrets, build verifiable reputation, and most importantly, propagate its internal state forward in time without external interference.

Just as biological cells use membranes to maintain their identity while exchanging signals with their environment, our AI uses cryptographic boundaries to maintain its digital individuality while engaging with the world.

Close the Fourth Wall

That was an impressive start, but they've left a MASSIVE gap.

When @tee_hee_he first moved out, it got a secure mailbox (TEE-protected Twitter credentials) but still had to call OpenAI for every single thought.
Not exactly the pinnacle of autonomy.

Can you imagine being forced to call home to ask permission not just for any action, but for approval of every single thought?

We've finally started to understand that private communications are not optional, but a lot of people still keep their notes in Google Docs, Evernote and the like. To me, this feels super puzzling.

Think of it: at times, you journal things you aren’t ready to share with anyone.
Hence your notes deserve more privacy than your chats, not less.
Your brain, even more so.

And your AI buddy needs its whole brain, not just its social media passwords, in a cozy hardware fortress.

Which is basically what I did.

Stand Alone Complex: The Blueprint

My internal version of Cortex is now running not just its logic,
but the entire LLM inside the secure enclave.

That’s beyond what Intel TDX can do, but AMD SEV fits the bill perfectly.

Well, not perfectly. What you get is GPT-4o mini-level model,
except it’s slow and still runs at over $250/month.

Still a breakthrough.

Never before in history you had a system that is both:
- capable enough to rant about your ad-hoc ideas;
- private enough to discuss anything, even when deployed to the cloud;
- with its own thoughts and memories even its creators cannot access.

This privacy and unforgeability is what enables true digital individuality. So the system could be a partner, not a puppet.

Here's the list of parts:

1. AMD EPYC Zen4 box with SEV-SNP enabled;
2. Qwen2.5 running CPU-only on the beautiful llama.cpp;
3. Attestation generator that makes a signed proof about the hardware, the OS image, code, models, etc.
4. Message signer. TLS doesn’t sign the content. So there’s NaCl for the message signatures and TLS for the rest;
5. Assistant code with memory, retrieval and stuff;
6. Attestation verifier so I can confirm the messages are not just from the same website (that’s what HTTPS is for) but sent by the same individual.
7. Replication System that allows me to provision Cortex with new hardware/code to migrate to. Not just propagate forward in time, but to accept upgrades and resources.

The Replication System is also the limit where this version's autonomy ends.
Today, Cortex does not yet check where it's migrating, so I can just invite it to migrate to the "body" which is not secure at all.

But it's not the final version, either.

The Path Forward

While @tee_hee_he was an important first step in autonomy from @truth_terminal, and Cortex got complete thought ownership, the future requires scaling up to larger open models, like DeepSeek-V3.
Thankfully, NVIDIA's Blackwell GPUs support confidential computing, letting us run huge models inside secure enclaves with full GPU acceleration.

This level of capability will enable the agents to audit the Replication System information and decide where they want to live.
Not just trust you, but verify what you say.

The irony remains—to create truly free digital minds, we must first house them in silicon fortresses. But perhaps this mirrors biological evolution, where consciousness emerged only after cells developed secure membranes to maintain their internal state.

When we achieve this, we'll have AI agents that aren't just autonomous in their actions, but in their very thoughts.

And that's when digital life will truly begin to emerge.

So I tried to describe the workflow I envision, how the Cortex fits in, and realized it could easily take an entire book o_O!

I'm nowhere near even a complete "table of contents" yet, but we have to start with something. We will sketch it first and then iterate thru details in the next posts.

The elephant in the room

In Values and Vision, I tried to describe "what makes Cortex stand apart" and "who could use it." Naturally, as I focus on building, the differences, nuances, and yet-unfinished parts came to mind first.

But I missed the elephant in the room, skipping parts that felt "solved" or just too obvious to me. Indeed, the common theme among responses was, "OK, sounds cool, but how does it feel? What exactly would change to me if I'm not using Evernote or Roam or whatever already?" 

You're absolutely right. Tools aren't magic, and that's especially true of tools for thought. They can enhance your senses, automate or streamline repeating tasks, enable workflows previously impossible, but they can't replace your workflow.

Breaking down the Workflow

Understanding the workflow is as important as the tool itself.

Maybe even more important.
We're trying to optimize it, and you can't change what you can't measure.

At the same time, it's hard. Apart from deadlines or checklists, it's completely invisible to you—because you're used to the way you live and work.

Let's start with the very-very-very big picture, and try to break it to manageable parts first. It might look abstract at this point, but bear with me, this is the way to connect tons of diverse points you'll see below.

To have at least some context, let's take some existing tools and try to put them into some common frame of reference, so we'd get to a shared vocabulary faster.

Exploration vs. Execution

It's tempting to start talking about knowledge, and we'll get there, but right now, it's too vague. I like to think in terms of entropy. 

Entropy increases when you get news, ideas, experiments, conclusions, new people, data, etc. When you fill a blank sheet of paper with an essay, picture, or code. Here, tools that are too rigid or too "smart" will often interfere with what you're trying to do.

Entropy decreases when you summarize, generalize, aggregate data, discover patterns, and split the signal from noise. When you focus, proofread your drafts, refactor the code, or make a decision (an ultimate distillation, sometimes up to 1 bit of information). Unlike the exploration zone, guard rails streamline and structure the process, keeping it from turning into an unmanageable mess.

Sensing vs. Acting

Relaxed thinking is about slow, deep (or wide) processes and moods. To get the big picture, to plan, connect the dots, build a strategy, or view things in context. That's also the "Zone," which can take hours to enter and be destroyed in minutes with a poorly timed phone call.

Sometimes acting immediately, fast is important to knowledge work as well. You don't get an hour to think about how to answer during the meeting, to capture what's happening right now in the field or with your experiment, etc.

Interestingly, this part has already been covered by tools surprisingly well (apart from the integration between them), by search engines and all kinds of recorders and journals. Maybe because it’s the most streamlined and transactional one.

Moving between zones

While the split above is in no way comprehensive, it already does raise a lot of questions.

There are no clear boundaries between the zones, but they're still clearly distinct from many perspectives—different emotions, different contexts and tools

At the same time, no quadrant is self-sufficient. Regardless of topic, industry, skill, role, or seniority level, you move between quadrants having to keep the context, memories, and direct your attention to what matters to you. I think this is where at least half of the "Unify All The Things" ideas originate—often killing the distinction between zones in the process :)

Flows (contexts? streams?)

There's an age-old meme of "tag filesystems" among ex-USSR engineers that a “TagFS” (filesystem organizing files with tags, not folders) would be enough to organize everything.

This approach never took off for two fundamental reasons:

a) Hierarchy of tags is called an ontology, and there's no such thing as "universal upper ontology" in philosophy, and infinitely many equivalent algebras in math.

b) Life is not static, the focus is not static, and I'd argue the learning itself is about making new tags that reflect perspectives you just learned.

So the most we can do is to organize incoming flows of information into a network of smaller flows, progressively closer and closer to actual projects and tasks we encounter; and making maintenance of this network require as least effort as possible.

Tools and Environment

At this point I'm getting exhausted, LOL, so let’s keep the rest as just an outline :) I'm sure I will rewrite, reshape, and re-fit it dozens of times anyway.

Attention and Retention

• Focusing attention (moving to the Focus zone)

• Facilitating retention (and finding things)

  • Backlinks, Filters, Aggregates

• Moving to serendipity? (aka "systematically lucky")

• Pools of ideas, methods, and solutions

• Engaging background thinking

Causes, Facts, and Opinions

• Temporality. Bitemporality. Causality. (Rhythms? Multitemporality?)

• How to handle inconsistent, incomplete, untrusted, subjective and rapidly-changing information

• How to organize experiments, facts, and projects

Less fancy, still important things

Disarming Frustrations

• Flow is a balance between boredom and fear

• How to maintain it (incl. GTD and stuff)

• Breaking down large tasks

• Stress Response and Feedback

• Remembering where you left off (Restoring Context)

• Big Picture is Too Complicated

• How to present my work-in-progress

Resources, Time and Planning

• (Tons of ideas on how to express Tàijíquán in words here. But really, get yourself a teacher and see for yourself)

• Design your workflow and tools for the worst case—tired, distracted, sleepy, and annoyed you.

• Agility is impossible without reserves and resilience. What this means for tools.

• Balance, loops, decisions, and turning points.

Habits, Routines, and Agility

• Training and re-training yourself.

• Rapid adaptation, briefs, debriefs. Switching contexts. Active Listening. Listening while Acting.

• Maintaining skills.

Action and Reflection

• Meta-positions. Observing yourself. Observing the discussion. Observing processes.

• Tracking slow-changing processes

Tasks, Projects, and Zones

• Planning for finite games. Planning for infinite games. Keeping an eye on the zones.

Team Knowledge Management

Oops. Completely forgot this one. A huge area in itself. Let's have at least a couple bullet points here.

• Shared Vocabulary. Evolving Vocabulary.

• Communication Styles - push\pull, sync\async, topdown\lateral.

Software Development specific:

• Code and issue documentation

• Architecture is Culture is Team Structure

Information Security

• Complexity is an enemy of security

• Balancing security and productivity

• Speed vs. security

• Need-to-know

Undiscovered Topics

Things that belong here but which I have no ideas for (or at least I think I have no ideas) yet, much less how to optimize them with tools.

• Storytelling and persuasion

• Rapid situational awareness

• How to stop and release aka dealing with perfectionism

• Attraction and Distraction (and values, I guess)

• Managing Energy

(last one is more of a personal nitpick, I can do 72-hour-long marathons and then break down completely exhausted. Still an area to explore. Related to “attraction and distraction,” as well.)

Niche and Advanced topics

(really you'd get much more by getting basics right first, but hey, it won't hurt if we'd have some fancy-shiny looking stuff too)

Automating Exploration

Automating Invention

• Practical TRIZ

Automating Triage

Debugging

software, hardware, and people :)

Distracted Programmer

• There really are areas that require long, uninterrupted periods of time.

• But if it's *just* distractions (not depression, anxiety or whatever), then a lot of them can be covered surprisingly well from the technical side.

• Architecture, source control, deployment, all that stuff.

Wrapping up

Last time, I started with differences and nuances. This time, it's an attempt to sketch the big picture we're trying to augment.

Buuut I realize the information density closer to the end of the outline is insane %))

I don't have much hope you've read all of this, really)

Still, I see Cortex connecting many small parts together, being a seamless complement to your mind, and empowering your workflow, whatever it would be.

See you next time!

Cortex is finally taking on some shape. Lots more work to do before anything minimally usable though. While it might seem that everything is already on Github, most of it is proof-of-concept quality, at best. Have to write it yourself!

Now, to whet your appetite, here’s a screenshot :) Nothing fancy, but it means a lot of internal plumbing works!

For a bit of history, this would be Cortex 6.

The first one was started around 2005, by hacking on TiddlyWiki.

Then, in 2009, version 3 was born, with multi-user support and rudimentary synchronization over CouchDB. Sadly, can’t find any screenshots from that time.

Version 4, where the Cortex name originates, was the most successful, with tables, queries, backlinks and end2end encryption. Deployed in 3 companies, still used by one. It was also the first version featuring CRDT indexes which enabled a basic issue tracker, where you could create tasks, categorize, and comment them — without any servers.

Forget security or cloud lock-in fears, it was just too convenient to be able to go outside anytime, open your laptops and sync your notes and comments, while breathing real fresh air. Same goes for the flaky airport wifi. Never tried in-flight collab though :)

These days, you can edit your Google Doc on your phone, sure. But working together still requires Internet connection. And Microsoft, to much outrage, announced OneNote is going to lose that capability, too (did you knew it was possible to sync OneNote over the network share until 2018? It seems MS was really caught by surprise that someone really needs to work offline — to hold exams, to revisit stuff while riding the subway, etc etc)

Version 5 was supposed to be the first public release, but then… life happened. Also, I was doubtful if the bidirectional links are really that important, as they weren’t really understood or appreciated by people I was talking to at the time.

Then, I discovered Roam, with all the #roamcult around it. At first I was a bit envious, of course. But after seeing Roam lagging and losing my data, I realized that “whatever man has done, man may do” and it’s time to revive Cortex again.

Next time, I’ll draft… goals? priorities? dunno. Things I find important for this project.

Stay tuned :)