Developing, Designing, Dreaming. Thinking. 20+ years in tech. And still spend most of my work hours on Google or Stack-overflow.
Coinbase Alumni. Ai @b3dotfun
100%. There's a clear difference between using an LLM w specialized tooling on top vs just an LLM directly. And Caddie does a great job of showing that ๐ค๐ป
In trading, stale data is just a slower way to be wrong.
- Caddie reads the market live the second you ask. ETF flows, liquidations, whale moves, fear & greed, all current, synthesized from every reliable source.
- But reading is just the start. same chat, you act on it: set a price alert, build a workflow, swap, hedge, automate the whole thing. The live data wires straight into onchain actions, wallets and gas already first-class on https://t.co/TqEOEGJfsh.
- General models are great at a lot of things. Staying live on the market is a different matter, and Caddie is built for it. right tool, right job.
Figuring out what a user actually wants when they're talking to your LLM agent is more of a PITA than you can imagine.
We spent months on this with Caddie, our AI workflow builder at @b3dotfun. Started with regex. Ended up with a 6-phase hybrid classifier:
1. Pending plan? (deterministic)
2. Last LLM suggested a plan or asked a question? (deterministic)
3. User asked a question with no workflow intent? (deterministic)
4. Editing an existing workflow? (deterministic)
5. Domain keywords or action verbs? (deterministic)
6. None matched? LLM fallback.
Our deterministic guards capture ~95% of intent. No latency, no tokens burned. The LLM only fires when we genuinely can't tell.
Here's a visual.
PS: Curious how others have handled intent classification. I'd imagine it's usually a hybrid. What LLMs downstream are you finding most performant for this specific use case?
Definitely one of the hardest challenges working w @b3dotfun on B3OS
- Due to LLM's being non-deterministic by nature, we had to ride a thin line between allowing our models to shine while at the same time not giving them enough rope to hang themselves with
- Training it to use specific domain knowledge we acquired over the years (especially while working @ Coinbase) helped a ton here
- In later iterations, we immediately introduced concepts like "recipes", pre/post processing & more. Which made Caddie much more deterministic and reliable, quickly making it something we felt confident trusting our $ in. Most if not all of B3OS employees have used Caddie directly to execute trades, for example.
AI agents are scaling fast; @METR_Evals reports their task completion horizon doubles every 7 months.
But there's a massive gap between an agent completing general software engineering tasks and an agent managing your onchain treasury. In crypto, where you're dealing with programmable money, "long running tasks" are often just a slower way to get exploited.
With B3OS's Caddie 1.1, we introduces a new standard: Verifiable Onchain Autonomy
๐๏ธHuman-in-the-loop: You approve the plan in plain text.
โ Deterministic Execution Engine: Every action is a trusted and battle tested explicit node
๐Absolute Control: Every input + result is logged, and wait nodes pause until you sign.
Hear our CTO Sean explain the architecture:
more hacks every day. I think the low-hanging fruit is old apps / contracts. newer ones I wouldn't be as worried about. might go to void, might get compromised, either way those approvals are still live
been cooking a workflow template on b3os for this. paste any wallet, scans 1000 txs across 12 mainnet chains, flags risky approvals to unverified contracts, gives you a risk score
sharing it for folks here https://t.co/EhVSwZ0cA3
stay safe out there y'all!
Great feedback & reception for the public beta of B3OS so far
No incidents / downtime, proves our execution layer is holding up well across SO MANY use cases (kudos to the eng team)
Now team is heads down on the next iteration: Agents
The agent primitive turns every B3OS user into a potential agent vendor, every action developer into a paid building-block publisher, and every consumer into a fee-tier participant.
There's essentially 3 layers of builders;
- infra (the B3 labs team)
- primitive builders (actions, triggers, x402 builders)
- agent builders (coming soon)
On the consumer side, agents & letting builders ship + monetize their own agent harnesses (all on an auditable execution layer) is gonna unlock some cool stuff of AI use cases
Turnkey wallets can now automate onchain + offchain actions.
With B3OS, @turnkeyhq wallets can execute, monitor, rebalance, pay, and notify across onchain apps and 2,000+ tools you already use, like Slack and Telegram.
That means you can:
๐ Mirror top traders on @Polymarket automatically
๐ Auto-rebalance yield on @Morpho
๐ Review payment approvals in Slack and settle onchain
๐ Sweep dust tokens into @USDC
๐ Fund wallets anonymously
Import your Turnkey wallet into B3OS, use our templates, or prompt your own workflow with Caddie AI or your own agent via MCP at https://t.co/4LDDuDeUMl
Today we're launching B3OS in public beta.
Onchain workflows any agent or human can run.
Live today:
๐ Traders copy-trading on Polymarket or trading X headlines
๐ Developers powering onchain apps with scheduled sends, recurring swaps, and more
๐ Teams paying vendors in USDC from email or Slack
Build in our UI or through our MCP server.
This is the future of autonomous onchain finance.
Life onchain will be very different after B3OS.
In the coming weeks we'll unveil B3OS: a product that began 10 months ago as a tool to ship our own products faster and more reliably.
B3OS has become our central focus and the operating system we use to innovate and improve our business.
We can't wait to show builders, traders, and enterprises what B3OS can do.
I had a call with the biggest gaming token on @base yesterday
They want to onboard people with
risk-to-earn, wages, real time prediction.
Iโm gamer since I remember but only a fool can continues to think games does need to be only fun, because people will always prefer sit in front of his PS5 and play Warzone or a single player path.
For the first time @b3dotfun wants to onboard creators from ALL AROUND the chains, because crypto gaming is an industry we need to grow together, not PvPing each others.
I will advise them helping on finding creators who really converts and not only my friends, itโs time to change the game.
If youโre hungry and you read all comment below.
Youโre part of that 1%