Hi guys, sorry to bother you, but I’ve got a super important job currently hanging for a DSL I’d really like to see what ends up with GPT-4.1-mini (approx. 590,000 records)
Can somebody bother support or something?
Job ID ==> ftjob-dX0NrZSbVwFcvt7curs7jHNV
@OpenAI_Support .. could you check what’s going on with Thomas’ hyperlambda model thing here?
Hyperlambda - Magic Cloud << you should probably also check out his stuff! It’s next level!
Thx mate, I think it worked, it started 3 hours ago 
Bookmarked, great stuff, thanks for sharing. Does magic allow describe and execute standard business operations? Sort of a file “here is how my business operates”?
That’s probably what the finetuning is for. Thomas is an exceptionally good developer. And it takes a lot for me to admit that.
Well, you can use AI to generate Hyperlambda. Hyperlambda is a DSL that does exceptionally well on database integrations (CRUD), HTTP web APIs, web scraping, headless browsers, etc, etc, etc. It’s an “AI agent orchestration framework” (whatever that is, definitions in this vertical seems to change every 5 minutes)
Think of it as an “API gateway” to allow Claude to (securely!!) access company resources. For instance, you can generate CRUD apps in seconds wrapping all tables with CRUD operations (no AI!), etc - As to “describe and execute standard business operations”, the answer is “depends upon your business operations”.
It’s not a silver bullet for everything AI, but if your needs are roughly similar to mine, it will definitely make you 10x more productive on anything related to creating “AI agents” …
The long term goal is to allow for it to completely replace stuff like N8N and LangChain with “natural language solutions” …
Hmm doesn’t n8n have natural language solution? Havn’t used it in a while though.
Natural Language Solution + control over the execution layer (whitelisting of function), is an area where I’m apparently 100% alone, minus WASI - However, that’s a different “product category”, since WASI requires compilation + deployment, equal to 5 to 15 minutes - Where Hyperlambda is generated (and often executed too) in 1 to 10 seconds (on average!)
The point isn’t “natural language to code”, it’s “natural language to secure code” …
Check out the whitelist keyword. It allows you to on a per function invocation layer specifically whitelist individual functions. Since everything is a function in HL, this gives you 100% control over the “execution surface”, on a “per individual request basis” - Which is why I’m so happily spreading around publicly URLs to shit allowing “random strangers” to write English into my sjait, while I generate server side code, I happily execute, and returns the result from - Simply put, because it’s not even possible in theory to prompt inject execution of “malicious code” - Such as this guy for instance, where I offer strangers $100 if they can hack it (by generating HL code from English that’s executed on my production server) …
It’s been up for 9 months now. So far, zero have collected “their money” …
… realising I’m probably the least popular guy on earth (according too Reddit at least), I’d say that’s an “above average achievement” for a tiny solopreneur, whom apparently is crazy BTW - Which I think you already knew a loooong time ago …
Anyways, so no, there’s nothing like this out there. Natural language to securely executed code, live, executed server-side, with RBAC integration, within a 100% secure whitelisted platform. Outside of Hyperlambda, these things simply doesn’t exist out there - And apparently, according to ChatGPT and Claude, it’s the single most important question of 2026 related to IT, AI and security
And then of course, there’s also the (obvious) issue, that “my LLM” happens to be 62.5 times faster than Claude, and delivers 10% of the bugs that Claude does … 
That’s not really too difficult haha