Introducing: VESSEL 🛥️
The thinnest, self-healing harness that lets you create perfect fat skills in minutes.
We got tired of watching agents fail at complex tasks after days of prompt tweaking, rate limits, and broken JSON.
So we removed the framework.
🟣 Zero friction — vessel create in 4 minutes
🟣 100% reliable — Pydantic + Tenacity + circuit breakers built-in
🟣 Thin harness — LLM only sends JSON, VESSEL guarantees the result
🟣 No Show more
You will never manually skillify an agent again.
Fully open source: https://t.co/bIJ7kyYgO3
WTF is a “loop” ? 😤
Well, if you’re not Boris or Peter, you probably don’t need to care
(Unless you have unlimited access to the models)
Because the “just put your agent in a loop” discourse is funny when you’re sitting inside a model company
For everyone else:
A loop = tokens
Tokens = money
Run enough loops and your API bill starts looking like a mortgage
So if you actually want agents running for hours or days, you basically have two options:
Build serious monitoring systems around token usage, budgets, limits, kill-switches, cost tracking, etc.
or
Use local models / cheaper models…
…but then congratulations, you’re now a part-time inference, optimization, and infrastructure engineer
The loop isn’t the hard part
Making it economically sustainable is
WTF is a “loop” ? 😤
Well, if you’re not Boris or Peter, you probably don’t need to care
(Unless you have unlimited access to the models)
Because the “just put your agent in a loop” discourse is funny when you’re sitting inside a model company
For everyone else:
A loop = tokens
Tokens = money
Run enough loops and your API bill starts looking like a mortgage
So if you actually want agents running for hours or days, you basically have two options:
Build serious monitoring systems around token usage, budgets, limits, kill-switches, cost tracking, etc.
or
Use local models / cheaper models…
…but then congratulations, you’re now a part-time inference, optimization, and infrastructure engineer
The loop isn’t the hard part
Making it economically sustainable is
Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use.
Its capabilities exceed those of any model we’ve ever made generally available.
Google's new algorithm just shrunk 31GB of memory down to 4GB 🤯
TurboVec is a new open-source tool that stores the data your AI app searches through, using 16x less memory.
→ 16x lower memory usage
→ Faster vector search
→ Works with LangChain & LlamaIndex
→ 100% open source
The race to build bigger AI models is loud.
The race to make them dramatically cheaper just got a lot more interesting.
Repo: https://t.co/udEIFLJGs6
Google's new algorithm just shrunk 31GB of memory down to 4GB 🤯
TurboVec is a new open-source tool that stores the data your AI app searches through, using 16x less memory.
→ 16x lower memory usage
→ Faster vector search
→ Works with LangChain & LlamaIndex
→ 100% open source
The race to build bigger AI models is loud.
The race to make them dramatically cheaper just got a lot more interesting.
Repo: https://t.co/udEIFLJGs6
This is CRAZY! 🚨🚨🚨🚨🚨🚨🚨🚨
Spent the whole day building a custom content creation skill for my Hermes agent, then asked him to generate a video promoting himself.
Now imagine combining this with your daily content !
This is CRAZY! 🚨🚨🚨🚨🚨🚨🚨🚨
Spent the whole day building a custom content creation skill for my Hermes agent, then asked him to generate a video promoting himself.
Now imagine combining this with your daily content !
Most people still underestimate how early we are with AI agents.
Right now feels a lot like the early days of SaaS.
The tooling is messy.
The workflows are imperfect.
Nobody fully agrees on best practices.
That’s exactly where the opportunity is.
I’ve been building AI-powered marketing, coding and autonomous Agents using Python, Claude, Playwright, LLMs, browser automation, and custom workflows.
For now, I actually prefer keeping parts of the process manual while validating real-world results.
But what fascinates me most is multi-agent systems:
• Agents exchanging information
• Memory vaults
• Parallel execution
• Autonomous decision-making
• Self-improving workflows
Very few companies have figured this out well yet.
The interesting part?
Businesses are already willing to pay significant amounts for systems that automate lead generation, outreach, SEO, research and marketing operations.
Curious:
What’s the most valuable workflow you’ve successfully automated so far ?
Most people still underestimate how early we are with AI agents.
Right now feels a lot like the early days of SaaS.
The tooling is messy.
The workflows are imperfect.
Nobody fully agrees on best practices.
That’s exactly where the opportunity is.
I’ve been building AI-powered marketing, coding and autonomous Agents using Python, Claude, Playwright, LLMs, browser automation, and custom workflows.
For now, I actually prefer keeping parts of the process manual while validating real-world results.
But what fascinates me most is multi-agent systems:
• Agents exchanging information
• Memory vaults
• Parallel execution
• Autonomous decision-making
• Self-improving workflows
Very few companies have figured this out well yet.
The interesting part?
Businesses are already willing to pay significant amounts for systems that automate lead generation, outreach, SEO, research and marketing operations.
Curious:
What’s the most valuable workflow you’ve successfully automated so far ?