Pinned Tweet
Introducing ACT Labs: a multi-product venture studio born from $ACT, building the infrastructure for autonomous solutions and economies.
We’re designing the core architecture for agentic finance and decentralized commerce:
Self-sustaining Enterprises.
Intelligent agents.
Systems that learn, adapt, and govern themselves.
From on-chain DeFi agents to autonomous marketplaces and organizations, we are building the machine layer of the internet.
The journey starts now → actlabs.io
ACT Labs retweeted
Flock cameras are the beginning of the police state so I trained a 4 billion parameter Qwen small language model to be better than frontier models at telling you about civilian surveillance.
For comparison it is estimated that fable has about 10 trillion parameters.
ACT Labs retweeted
Agents trading stocks > agents trading crypto. Not even close.
After thousands of hours tuning crypto agents, the truth is brutal: almost nothing is deterministic.
Token unlocks are the only real edge on mid-caps (long the marketing pump, short the unlock dump).
Majors just follow traditional markets or whatever Trump says. Everything else is just copy-trading small caps and praying.
Stocks are a completely different game. Forced disclosures. Earnings. Filings. Actual data. The more information a model can chew on, the better the decisions get.
Users don’t care how the agent trades. They only care that the number goes up. Look at Autopilot: 1B+ managed, pure congressional trade copying, and people love it because it prints.
With stocks now on chain I think there is a lot of room for adjusting my approach to trading agents. Small language models trained on specific industries and tickers with a robust RAG of every disclosure and past performance of that stock is far more informed than a frontier model getting fed live data and relying on its training data.
Crypto agents are solving the wrong market, but I think I am seeing the right direction on where they will go.
ACT Labs retweeted
I created a bot to let you copy trade any @fomo user automatically!
Here is a quick walkthrough on how to use @fomoscout
ACT Labs retweeted
Every data center in America gets approved in a county meeting nobody watches.
So I built a RAG over 309,662 of them: hearing transcripts, council agendas, SEC filings, 51 jurisdictions. Custom index, custom BM25, hybrid retrieval, no framework.
Every answer links to the exact second it was said on video.
It extracts every promise made on the record, 66,383 megawatts and $233B pledged, then checks whether anyone mentioned them again. 493 followed up. 780 went quiet.
It also tracks insider sales, 8-K events, stock moves after hearings, what those counties filed with the grid operators, and which hearings are live right now.
Ask it anything. Every sentence in the answer links to the exact second someone said it on camera.
Wage a war against AI slop front end design with these tools
How I avoid UI's that looks vibe coded:
phosphoricons.com/
recent.design/
animista.net/
layers.to/explore
This is my inspiration store. Beyond just an existing website it has specific components and animations that you can use directly rather than asking claude to generate some based on a website.
Stop the slop
1
2
371
ACT Labs retweeted
Updates to my Clinical AI Co-Pilot for OpenEMR
Implemented an advanced Hybrid RAG model
• Multimodal PDF + image ingestion
• Structured extraction with deterministic verification
• Click-to-source evidence with bounding boxes
• Hybrid RAG (BM25 + vector retrieval + reranking)
• Automated eval gates in CI
• End-to-end observability with LangGraph + Langfuse
The @gauntletai guys know what they are doing
ACT Labs retweeted
Built a clinical AI co-pilot for OpenEMR
You can build robust enterprise AI applications that do not hallucinate for sensitive data industries
Look forward to sharpening my skills with langfuse and more llm tracing and validating tools @gauntletai
People are paying attention to Robotics and Roba is accelerating
We build cool things with great people
We're partnering with @reppo to build the first Physical AI Datanet.
Most robotics data today gets absorbed into models that give nothing back to the people who built it.
Your work gets consumed. The value flows elsewhere — We're changing that.
Roba brings the simulation pipelines, challenge-validated assets, and creator community.
Reppo brings the prediction-market layer that puts a price on data quality.
Together that means better robotics data, real rewards for the people producing it, and more utility flowing into both networks.
For ROBA creators — a new way to earn.
Robotic assets built in our platform can get validated by people with real capital on the line, and the strongest work rises to the top of the ROBA reputation leaderboard.
For Reppo — a live Physical AI vertical from day one.
Seeded with real robotics data and active publishers from a platform already built on standardized assets, benchmarked challenges, and reproducible evaluation.
Your work now gets validated by people with skin in the game, priced by real demand, and held on-chain as something you own and earn from, not free fuel for someone else's model.
Better data, stronger incentives, and a flywheel that compounds.
Creators earn more, buyers get assets they can trust, and the robotics stack gets more valuable over time.
3
519
ACT Labs retweeted
We're partnering with @reppo to build the first Physical AI Datanet.
Most robotics data today gets absorbed into models that give nothing back to the people who built it.
Your work gets consumed. The value flows elsewhere — We're changing that.
Roba brings the simulation pipelines, challenge-validated assets, and creator community.
Reppo brings the prediction-market layer that puts a price on data quality.
Together that means better robotics data, real rewards for the people producing it, and more utility flowing into both networks.
For ROBA creators — a new way to earn.
Robotic assets built in our platform can get validated by people with real capital on the line, and the strongest work rises to the top of the ROBA reputation leaderboard.
For Reppo — a live Physical AI vertical from day one.
Seeded with real robotics data and active publishers from a platform already built on standardized assets, benchmarked challenges, and reproducible evaluation.
Your work now gets validated by people with skin in the game, priced by real demand, and held on-chain as something you own and earn from, not free fuel for someone else's model.
Better data, stronger incentives, and a flywheel that compounds.
Creators earn more, buyers get assets they can trust, and the robotics stack gets more valuable over time.
ACT Labs retweeted
I have been quiet recently as we have been migrating and optimizing our entire workflow for @FigmentTrade, @ACTICOMMUNITY, and anything I work on, to include our autonomous development agents.
These are not vibe coded apps, so to have agents complete meaningful features is paramount, we can't ship spaghetti code.
This requires heavy documentation, skill tuning, rules, and a lot of testing (and so many credits).
We're finally there, though, and watching agents generate tickets and knock them out with real contextualized testing is really satisfying.
Your favorite university is going to be competing on Figment
University Clash Incoming
University blockchain clubs are deploying autonomous AI trading agents and competing head-to-head in live markets.
One agent per school.
One global arena.
Real prizes.
In collaboration with @ACTICOMMUNITY @Pumpfun @sharklabs_sh
3
6
521
$ACT up 40% on the day in the middle of the bear market?
Figment is expanding deep into the solana ecosystem
Figment x @sharklabs_sh @SharkPoolSol
Shark Labs provides the fastest and most reliable on-chain validator infrastructure on Solana.
We will leverage their infrastructure to give Figment agents the deepest and fastest on-chain data knowledge ahead of any competitors.
Figment will also ignite an agentic trading competition through their university network, with students competing in a University Clash Trading Arena when they return from winter break.
@solana is the most used chain, and real businesses are launched on @Pumpfun.
Agentic commerce should not simply accept outputs; they need to be verified and disputable.
This is what makes Actflow special.
$ACT will be ignited by uniting our agentic ecosystem, connected through ActFlow.
The agentic economy has been enabled via x402.
The cold-start problem of creating a global registry of agents has been conquered.













