Sohejl

AI, Web3, Strategy | Founder of @TEACHMEDEFI | Previously Web3 & M&A Advisory at BDO.

[Sohejl @sohejl_]

Without privacy, AI becomes surveillance.. Throwback to Nov, when I got to join a panel on AI, Privacy & Web3 in London at the @MidnightNtwrk Summit. An amazing venue to be discussing tech at ->>

The Old Royal Naval College is a UNESCO world heritage site. It’s a historic technology & science hub. The home of the Prime Meridian and Greenwich Mean Time (GMT)!

I shared some thoughts on how Web3 can help AI go beyond blind trust.
AI is obviously super useful and increasingly so in many contexts. But it’s also a black box.
We should be able to rely on credible guarantees that our data is treated confidentially and that we are receiving the exact service we are expecting.

Legitimate questions that come up for any person or company using AI:

What happens to my data? Is it kept private? Who says so? How can I verify that? Is the output actually coming from the AI Model I expect to be using? …or from an AI model at all? How do I know for sure?

This is what we are working on at @OODA_AI_ Full video linked in comments.

Thanks for having us @MidnightNtwrk

[Sohejl @sohejl_]

Using OpenClaw just got 19x more expensive if you use Claude. But you can fix this.

If you've been running Claude through a third-party tool like OpenClaw your subscription just stopped covering it and you probably got a message from your agent telling you this like I did.

Pro ($20/mo), Max ($100-200/mo): OAuth tokens are now restricted to Anthropic's own products. Everything else moved to pay-as-you-go API pricing effective April 4.

I run my personal AI agent on OpenClaw and did the math:

Claude Sonnet 4.6 on the API costs $3 per million input tokens and $15 per million output tokens. For an agentic workflow with lots of back and forth, context retrieval, and multiple calls per task…that can add up quickly :)

when subscriptions worked in third-party tools, the effective rate was way lower than API pricing.

On one case a developer ran ~10 billion tokens through Claude over 8 months on a Max plan for $800 total. The same volume at API rates would have cost over $15,000. That‘s a 19x difference.

Other reports vary but a cost difference of more than 90% is mentioned pretty much regularly.

As part of the switch, Anthropic gave a one-off credit for the pay-per-use API in the amount of your monthly subscription price as credits. I already used that up within a day! Shows you the huge difference in number of tokens you get for the same price.

You can look at alternative open-source models, that can compete in quality, to counter this cost increase.

Claude‘s Sonnet model costs 3$ / 15$ on the API (per million input/output tokens). You could use a competitive open source model through eg. Openrouter like: MiniMax M2.5 (0.1$ / 1$ ) or GLM-5 (0.7$ / 2.3$ ) or DeepSeek V3.2 (0.26$ / 0.38$ )

Either one is multiples cheaper. This makes open-source models look a LOT more attractive if you‘re using AI agents.

Just send this post to your agent and tell him you want to make this change and to tell you the steps. Easy.

[Sohejl @sohejl_]

490 open browser tabs. Most people just forget about them. I turned them into a searchable knowledge base in 20 minutes.

@itsolelehmann wrote about @karpathy‘s viral LLM knowledge base post today and ended it with "someone please build this."

Karpathy's idea was to take everything you're interested in (articles, papers, tweets, videos etc.) dump it into one folder, point your AI at it, and let it read, organize, and index everything. Then you can query your entire personal library with natural language. Ask it to connect ideas across sources, surface things you forgot you saved or summarize everything you got on a topic. And the AI maintains it all as you add new stuff.

I had 490 open Safari tabs. Basically an unorganized dump of links to read later. Which I never end up reading later anyway :)

So I sent Karpathy's post to my AI agent and after some back & forth, it actually built this in one short session
I pasted all 490 URLs. I asked Pulse (my AI agent) how to do this easily and it told me how to easily bulk copy links in Safari on your phone with 2 clicks.

Out of the 490, it filtered to 251 worth saving: 206 X posts, 21 articles, 15 tools, 6 GitHub repos, 3 research papers. The rest was skipped automatically. All categorized, frontmatter written, saved to the right folder. Done in under 20 minutes!

It pulled metadata on every X link. We ran into a few bugs which got fixed after some back and forth: most of the "tweets" weren't tweets. They were full X articles sitting inside empty post wrappers, invisible unless you know where to look in the API response. 158 of 206 posts turned out to be complete long-form articles that were missed. It went back and reextracted it all.

What the system does now (image attached):

I drop any URL into the chat. It gets filed, tagged, indexed. Zero effort on my end. The ongoing maintenance is basically free.

I query it naturally. "That David Deutsch article about raising children" -> finds it. "Connect new ideas and frequent tips regarding SEO across my knowledge base" -> synthesizes it across 30+ sources. "Something about subscription pricing in a newsletter I read a while back" -> semantic search gets it from a half-memory.

Now and then, it surfaces relevant material proactively without me asking. The attached image was also generated by the agent itself, showing how the full flow works. The rabbit hole honestly doesn't end. You can do so much now, simply with an idea + giving an AI agent the right tools + iterating on the system together. It's wild.

We‘ve never had such high leverage on human creativity ever. As usual, technology keeps shifting the arena from the perspiration phase to the inspiration phase. We don‘t have to do the arduous tasks. „1% inspiration and 99% perspiration“ is a misleading idea of how progress happens: the perspiration phase can be automated.

I think creativity & understanding will increasingly become differentiators. And it has never been easier to build both: by using AI.

(Sidenote: I already have a personal AI agent running 24/7 on a server, with access to obsidian notes and QMD for semantic search. So the plumbing for this use case was already in place. But this setup is too much to get into for this post…will get into it another time!)

[Ole Lehmann @itsolelehmann]

karpathy just casually described the future of ai and most people scrolled right past it:

he's been building what he calls "llm knowledge bases."

here's what that means in plain english:

you take everything you're interested in.

articles, research papers, datasets, images, etc and you dump it all into one folder

then you point your ai at the folder and say "read all of this, organize it, and remember it"

the ai reads through every single source.

writes summaries, groups related ideas together, links concepts across different articles

basically builds a personal library that's fully organized and searchable

and it maintains the whole thing for you.

when you add something new, the ai reads it, figures out how it connects to everything already in the library, and updates automatically.

karpathy said he rarely touches it himself

once the library gets big enough (~100 articles, ~400k words), you can start asking it complex questions and get answers pulled from across your entire collection

> "what are the common themes across these 30 papers"
> "what did i save six months ago that connects to this new idea"
> "summarize everything i have on topic x and tell me what's missing"

and every answer it gives gets filed back into the library. so the system gets smarter every single time you use it.

the memory grows from both sides: what you save AND what you ask

now think about your own life for a second

you probably have

> thousands of twitter bookmarks you'll never reopen.
> hundreds of saved articles from the last year
> podcasts where someone said something brilliant and you can't remember what it was or which episode

all dead knowledge.

you consumed it once and it disappeared

now imagine all of it lives in one system: organized, connected, and queryable.

you could ask "what are the best pricing frameworks i've come across this year" and get an answer that pulls from:

1. a podcast you listened to in january
2. a twitter thread you bookmarked in march
3. and a blog post you forgot you even read

the ai connects dots across formats, across months, across topics.

because it absorbed everything and has photographic memory of all of it

that's the dream. and karpathy built it

the problem: right now this requires obsidian (a note-taking app built around linked notes), command line tools, custom scripts, and browser extensions just to wire it all together.

you need to be quite technical

karpathy even said it himself: "i think there is room here for an incredible new product instead of a hacky collection of scripts"

i think whoever packages this for normal people is sitting on something massive.

one app that syncs with the tools you already use, your bookmarks, your read-later app, your podcast app, your saved threads.

it pulls everything in automatically, the ai organizes and connects it over time, and you can ask questions across your entire personal library whenever you want

you never manually upload anything. it just learns in the background

someone please build this

[Alex Houdz 🍉 @houdz_kek]

Happy friday everyone :)

Alright, I'm jumping in the big unknown with you guys!

For years i've watched from the side while builders were in the field, showing their demo, their metrics, wins and fails.

Confronting my idea to the world is scary.
But I decided it was enough, i can't hide behind anything now that vibe coding is there.

So here's my pitch to you guys, hopefully this resonates with you 🫶

3 days ago I started setting up OpenClaw

i'm not scared with technical things, I like watching tutorials. But still, I got frustrated setting it up. A lot of trial and error.

Then i was shocked when I saw how much a single prompt costs on API compared to a normal chat plan.

All this time, i kept thinking about the conversation I had with my dad last Monday, trying to convince him to try Claude Pro.

And then the idea came in like it was evident.

OpenClaw, (which is arguably the next gen of the best of AI) is completely inaccessible to non-tech people.

The setup is not smooth. The terms are scary: VPS, secure gateway, API credits, MCP servers. Normal people close the tab in 3 seconds.

And here's the thing.
I've seen this exact same pattern for years in crypto.

Everything was so technical, most people never try the apps.

So we started doing "abstraction". You abstract the tech from the user.

This is what I'm building.
Abstract the setup and the jargon from OpenClaw, so the rest of non-tech people can use it where they already are: WhatsApp.

No terminal. No config files. Just a WhatsApp conversation.

You chat with Pepe. He sets up you up. Calendar, email, reminders. You control what he can see. Toggle on, toggle off.

We started the building MVP.

If you want to be an early tester, DM me or check the link in comments 🥰

[Sohejl @sohejl_]

Full panel video: https://t.co/mQqqbTEKg2

[Sohejl @sohejl_]

Without privacy, AI becomes surveillance.. Throwback to Nov, when I got to join a panel on AI, Privacy & Web3 in London at the @MidnightNtwrk Summit. An amazing venue to be discussing tech at ->>

The Old Royal Naval College is a UNESCO world heritage site. It’s a historic technology & science hub. The home of the Prime Meridian and Greenwich Mean Time (GMT)!

I shared some thoughts on how Web3 can help AI go beyond blind trust.
AI is obviously super useful and increasingly so in many contexts. But it’s also a black box.
We should be able to rely on credible guarantees that our data is treated confidentially and that we are receiving the exact service we are expecting.

Legitimate questions that come up for any person or company using AI:

What happens to my data? Is it kept private? Who says so? How can I verify that? Is the output actually coming from the AI Model I expect to be using? …or from an AI model at all? How do I know for sure?

This is what we are working on at @OODA_AI_ Full video linked in comments.

Thanks for having us @MidnightNtwrk

[Sohejl @sohejl_]

@rene05x @TEACHMEDEFI @DJohnstonEC @MorpheusAIs Glad you enjoyed it and i hear you! :)

[Sohejl @sohejl_]

@DJohnstonEC @TEACHMEDEFI @MorpheusAIs Was lovely having you on the show, mate!

[TEACHMEDEFI @TEACHMEDEFI]

Episode 116 (!) of TEACHMEDEFI features @DJohnstonEC and @sohejl_ chatting about @MorpheusAIs

- vision for the "agentic age"
- "not your keys, not your AI" crypto corollary
- "Ghost Protocol" HTTP "Payment Required" Code 402
- x402, Agent Kits, machine-to-machine commerce

Did you know @AskVenice uses Morpheus? Tons to discover in this fascinating conversation with David.

Timestamps:
0:00 - AWS Outage & Decentralized Infrastructure Resilience
2:57 - AI Aggregator Model vs Direct GPU Operations
8:00 - Inference Provider Integration & Venice Partnership
11:03 - MCP Tool Integration & Agent Reputation
14:04 - Private Data Access & Database Limitations
18:07 - AI Clone Examples & Base Integration
21:03 - ChatGPT Privacy Concerns & Government Access
24:08 - Peer-to-Peer Routing & Light Client Strategy
29:00 - Fair Launch Model & Community Structure
32:14 - Protocol-Owned Liquidity & Token Mechanics
38:16 - Long-Term Project Vision & Sustainability
43:13 - Token Allocation & Daily Emission Distribution
45:49 - Project Milestones & Current Development Stage
48:08 - Historical Context & Fair Launch Origins
51:38 - Personal AI Ownership Call to Action

[Sohejl @sohejl_]

@houdz_kek @thefintechtimes You!

[Sohejl @sohejl_]

Thank you for the shoutout, The Fintech Times. The “trust me bro” paradigm is a real problem 🥹

From my panel discussion at the Midnight Summit 2025.

[Sohejl @sohejl_]

Link to article: https://t.co/HxCG29cZgX

[Sohejl @sohejl_]

Thank you for having me @outlierventures and @midnightfdn!

Was great to share our story at @OODA_AI_ and what‘s next.

See you soon on @base…

[OODA AI @OODA_AI_]

Proud to have our story shared by our partners at @outlierventures and @midnightfdn.

Our Chief Web3 Officer @sohejl_ presented the journey of our Founder & CEO @charl1e_ and unveiled our vision for launching on @Base.

[Sohejl @sohejl_]

We are joining The Post Web Base Camp run by our investor and partner Outlier Ventures! @OVioHQ

Together with our partner @MidnightNtwrk, we, @OODA_AI_ are exploring how to bring more privacy to the use of AI. We are using zero-knowledge proofs to verify that an inference was indeed carried out in a secure, confidential environment.

What does that mean?!

That means when you use AI, such as an LLM, you can be sure your data is processed confidentially.

Through credible guarantees, not blind trust!

Had great fun bringing this to life with the team and our partners.

Learning a lot as we go. A daily reminder that problems are soluble! 😅

[OODA AI @OODA_AI_]

Excited to be building with our partners @OVioHQ and @MidnightNtwrk as part of the Post Web Base Camp Cohort!

With Midnight, we’re exploring how zero-knowledge proofs can be applied to verified AI inference on decentralized networks, bringing trust + privacy to AI at scale.

Big thanks to our investor and partner Outlier Ventures and congrats to all the amazing teams in this cohort!

[Sohejl @sohejl_]

AI that everyone can access. That anyone can verify. That no one can control.

That’s what we’re building with @OODA_AI_ and I‘m excited we got a new partner with @PhalaNetwork on that journey.

Phala brings a strong track record and expertise when it comes to TEE-secured compute. We’re combining that with our work on decentralized inference + zero-knowledge proofs (with our partner @MidnightNtwrk) and our pragmatic real-world AI solutions used by large enterprises across Europe today.

This gets us: AI that’s not only smart, but provably honest and private.

No middlemen. No unverifiable black boxes. No blind trust in centralized AI.

Instead, we get verifiable intelligence, running on secure hardware, backed by cryptographic proofs.

We’re a Nasdaq-listed company ($OODA) shipping real-world AI solutions and infrastructure today — and launching a utility token soon so anyone can plug in, contribute, and earn.

AI shouldn’t be the domain of a few labs. It should be a universal resource — powered by the many, not just for the many.

Let’s make that happen.
Universal AI is coming.

[OODA AI @OODA_AI_]

We’re excited to partner with @PhalaNetwork to advance trustless, verifiable, and powerful AI — built securely with TEE technology.

OODA AI is proud to be the first Nasdaq-listed company launching a utility token, enabling a future where AI isn’t just open to all — but powered by all.

We’re at the intersection of practical AI applications, APIs, agents, and tools — where Web2 user experience meets Web3 resilience. Bringing secure, decentralized AI to the masses.

You can already join our journey via traditional finance by investing in $OODA on Nasdaq — and soon, through our upcoming utility token.

The future of AI is accessible, verifiable, and decentralized. This is what we call Universal AI.

[Phala @PhalaNetwork]

Partnership Announcement: Phala 🤝 OODA AI

We’re joining forces with @OODA_AI_, a Nasdaq-listed AI powerhouse trading under ticker $OODA, focused on AI Apps, API:s & Agents, GDPR-grade data security, and decentralized AI computing.

Together, we’ll advance safe AGI via TEE-secured AI for both CPU & GPU compute.

[TEACHMEDEFI @TEACHMEDEFI]

https://t.co/BXfayulEms

[TEACHMEDEFI @TEACHMEDEFI]

Podcast bros 🥹 #ETHDenver2025

[Sohejl @sohejl_]

🔥🔥🔥

[TEACHMEDEFI @TEACHMEDEFI]

@Polkadot booth going strong at ETHDenver! 🔥🔥🔥

[Sohejl @sohejl_]

@GldnCalf @The_Weak_Hand 🧨

[Sohejl @sohejl_]

@LearnPolkadot @The_Weak_Hand @TEACHMEDEFI @Polkadot @TheKusamarian 🥳🥳🥳🥳