Matt DeCelles

@PrestonPysh Thank you for all that you did Preston! You were instrumental in my journey and I am glad that we were able to meet up a few times over the years. Godspeed.

A personal note.  After years of public discussion/work: the podcast, social media, venture capital, everything — I'm stepping back to focus on my family. My kids are growing fast.  My wife deserves the best of me.  To everyone who listened, read, grew, and built alongside me — thank you. What a blessing you all have been!

@mikealfred People just don't understand how big a deal this is. Will take time of course but man... huge.

@Werkman congrats man! You guys are crushing it.

👀 $SATA

@btc_overflow @FMTadvisory Can you let us know when you buy your leaps. Asking for a friend. :)

@FMTadvisory Yep. It's encouraging to see how dead it is to bad news. Few months of sideways grind to flatten IV and positioning could get very fun.

Bonfire in markets. Bitcoin only down 0.2% today.

@bradmillscan @Voxyz_ai Codex 5.4

@Voxyz_ai how do you reprogram the openclaw agent to stop reading the memory files and read the new system

my 5 agents store memory in markdown files. one reading is fine. five concurrent reads and writes, lost data, overwrites, conflicts. hit all of them. i wrote "check memory before answering" in the prompt. retrieval rate: 30%. reminders don't work. the filesystem was never a database. we just used it because nothing better existed. the fix: Deeplake. each agent gets its own postgres. no ops, done when the job is done. the surprise: it's not just structured data. agents can store and search images, video, and PDFs in the same system. one memory layer for everything. no more 5 agents fighting over one markdown file. each agent gets its own space, its own data, its own boundary. the memory layer for agents is due for an upgrade.

@bradmillscan @huntharo @openclaw its going to be so much better than what you are doing now.

@huntharo @openclaw How dangerous is this for people like me - non coders who don’t vibe code and don’t use Claude code or gpt codex for coding, but do use their openclaw to code things? Oh and who hit edge case bugs constantly

/codex_plan mode in @openclaw via Codex App Server integration. Plan preview in message. Full plan in markdown attachment. Can ask and answer steering questions too!

@melvynx @bradmillscan Get this tattooed.

OpenClaw is not for coding. OpenClaw is not for coding. OpenClaw is not for coding. OpenClaw is not for coding. OpenClaw is not for coding.OpenClaw is not for coding.

@hillery_dan No strk?

I still think if you are an institution, STRD is the best hold here.

@MatthewBerman @agentmail We built a tool for perfect memory. 100% perfect recall but its very new and oriented towards enterprise. But would love to show it to you if interested.

Who is building products and services for agents? @agentmail and here.now are super interesting. Who else?

@theo Any Updates on T3 Code. loving it so far!

Live in a bit. Any news drop that I need to cover?

@IIICapital This is the new narrative for this cycle.

1. Risk on? Bid flows to BTC and Amplified BTC. 2. Risk off? Bid flows to Digital Credit, which flows to BTC. Has the market converged on a perma-bid for BTC?

Macrohard or Digital Optimus is a joint xAI-Tesla project, coming as part of Tesla’s investment agreement with xAI. Grok is the master conductor/navigator with deep understanding of the world to direct digital Optimus, which is processing and actioning the past 5 secs of real-time computer screen video and keyboard/mouse actions. Grok is like a much more advanced and sophisticated version of turn-by-turn navigation software. You can think of it as Digital Optimus AI being System 1 (instinctive part of the mind) and Grok being System 2. (thinking part of the mind). This will run very competitively on the super low cost Tesla AI4 ($650) paired with relatively frugal use of the much more expensive xAI Nvidia hardware. And it will be the only real-time smart AI system. This is a big deal. In principle, it is capable of emulating the function of entire companies. That is why the program is called MACROHARD, a funny reference to Microsoft. No other company can yet do this.

@bradmillscan Hey Brad my buddy also has fully solved memory for agents. 100% perfect recall. Zero drift, validated across 4,800+ memory objects and adversarial attack suites. Were looking for a few people to beta it to get feedback.

power went out last night ... my openclaw came back completely regressed from all the progress we made yesterday and the day before. Forgets everything. What could have possibly happened...he's back to being a real knob.

Hey Brad my buddy also has fully solved memory for agents. 100% perfect recall. Zero drift, validated across 4,800+ memory objects and adversarial attack suites. Would either of you or @jlehman_ be intersted in learning about it. Were looking for a few people to beta it to get feedback.

@mattdecelles 💪 cc @jlehman_

Unlocked a lot of capabilities today. Now we’re using better compaction process with a DAG … One session that goes for weeks instead of hours = more context Semantic memory To search workspace files = better memory ACP where sene is orchestrator and PM getting codex to do the coding work = better multitasking Isolated agent with its own workspace and bootstrap files that sene controls, setup for my wife with bluebubbles for her to talk to over iMessage = more functionality We also have Gateway Level visibility with Opik logging everything Spent a lot of time today using ACP to make major progress on a big project to merge all of the previous 28 days of sessions into the live lossless-claw DAG - when we’re done we’ll ship the code to the lossless claw project as a PR cc @steipete @jlehman_ @cathrynlavery 🙏

This is the biggest story in finance. Yet few are aware what's happening.

@Scobleizer I've been using notebook LM a lot to learn things super fast. Smart to delpoy X api to ingest topics.

AI can read X better than you can. Then it can create a Notebook LM for you. Here's today's news, as a podcast, a slide deck, a mind map, video to come in a bit: notebooklm.google.com/notebook/50ab4… All gathered by Levangie Labs by reading tens of thousands of posts here on X through the X API.

@Scobleizer wow this is soooo cool!

@fejau_inc @Mattmaximo @thsottiaux this is the way

@Mattmaximo @thsottiaux i've found the best setup atm is openclaw running opus 4.6 that delegates coding tasks to codex CLI running gpt 5.4

Codex's biggest flaw continues to be its poor communication skills Claude is so much better to talk to

@bradmillscan @Voxyz_ai @vincent_koc great job man!

Day 28 with OpenClaw ... I'm finally making some serious progress over the last few days. Moved into the Mac Studio apartment (and posted a guide) Reworked the bootstrap files. Semantic Memory working (@Voxyz_ai) Got the Opik gateway plugin going for visibility (@vincent_koc) Switch to using lossless-claw DAG (@jlehman_) ACP working with Codex as persistent agent 2 bug reports got real traction (& my agent got a GitHub shoutout) ... getting my head above water finally..