Levanto Labs

Try Sage today with 100 calls for free. We'd like to hear what you're using it to build, or what you wish it was better at. DM for more credits!

@goinfrexeth Cheers

@levantolabs Congrats!

Meet the first AI to say ''I don't know''. We think agent mass adoption is blocked by three things: 1) weak security, 2) unreliability, and 3) how hard agents are to set up. Our first product, Sage, is focused on reliability. It's a "decision model" - a safer, faster, and cheaper way for machines to choose, act, and escalate to a human when confidence is low. You give it content (up to 32K tokens) and a list of questions (Sort, Yes/No, Choice, Tags, Scale), and Sage answers in 200ms - 9x faster than a traditional LLM - always with a confidence score attached. So yes… it's humble enough to say "I don't know." You can also turn on "grounding" to automatically run a web search and enrich the context. Under the hood: we took an open-weights LLM and fused on a classifier through post-training. It's great for agentic workflows, agentic guardrails, data pipelines, content moderation, operations, and risk & fraud. Why does this matter? Today's LLMs are great for chatbots, research, and creativity - but automation needs something much faster, with structured outputs, that isn't overconfident and is ready to admit when the signal is too weak. Sage preview is live. Excited to see your feedback. Levanto Labs is out of stealth today, founded by @marco_derossi and @bigironchris. We are hiring, reach out! Check the links in the post below 😊

@CreaoAI You send us an angry DM, and we make the model better.

@levantolabs When it gets something wrong and the agent still proceeds, what's the recovery path?

@pengmilio @bigironchris Cheers

@levantolabs ok this is actually cool good stuff @bigironchris

@BellingBird Let us know how you like it.

@levantolabs I'll be checking out the preview, if this works as advertised, you're solving the exact headache that keeps me from trusting most agents with real work.

@0xYuriiB i will promise you

@levantolabs Wishing this young and promising man the best of luck!

@EXM7777 be an LLM -> dont know the answer -> guess and sound really confident -> human acts on it We need to change this.

an OpenAI researcher basically admitted that models are trained to make things up... it's just the way they're built tho: during training the model is graded against correct answers, and correct answers cite sources so when it hits something it doesn't know, "i don't know" scores zero... inventing a source that looks right scores full marks repeat that millions of times and you get a model that fabricates before it admits uncertainty and the behavior spreads everywhere... a model that can't tell what it knows will bluff on anything, citing a paper or approving a refund, same move that's the actual ceiling on autonomous agents right now intelligence isn't the missing piece... you can't let an agent act alone if it never says "i'm not sure" the fix is a model that outputs a decision with a confidence score and pulls a human in when the score is low that's what @levantolabs is building... an AI that knows when to say "i don't know" is worth more than a smarter one that never does

@zaimiri We built this for us, and hope others find it useful. Check it out for free today levanto.ai

I spend a crazy amount of time building AI systems. The most annoying thing is how confidently they are wrong sometimes. LLMs are still too agreeable. They can't say "I don't know" or "Not sure". Sage’s calibrated confidence gives agents a point to stop and pull in a human (if needed). Excited to test this against some of my own workflows to see where my models are wrong but pretend to be right.

@M1ttelmeer @zaimiri Well that's up to you basically. That's where you'd do a review.

@zaimiri the "i dont know" output is honestly the feature i want most from every model right now curious how sage handles edge cases where the confidence score is mid but the answer is still right

@withhimu @marco_derossi We think there's a few big use cases were this is going to go well.

This looks absolutely amazing at many levels. First of all I admire the work done by @marco_derossi for dAI ecosystem. But also, this looks like something a traditional business can use to solve real problem. I beleive a decision model is similar to 'Model-as-a-Judge' kinda system that offers a layer of objectivity instead of compliance and user-pleasing models often fall for.

@0x_kaize @marco_derossi @bigironchris THE DECISION MODEL IS LIVE.

AI startup just launched a model whose MAIN FEATURE is admitting "i don't know". Levanto Labs came out of stealth today ( founded by @marco_derossi and @bigironchris ) with Sage - a "DECISION MODEL" built for one job: - Not a chatbot - Not another frontier LLM - Model that lets machines choose, act, or escalate to a human when confidence is low How it works: You feed it content ( up to 32K tokens ) + a list of questions ( sort, yes/no, choice, tags, scale ). It answers in ~200ms ( 9x faster than a traditional LLM ) Every answer comes with a confidence score attached. If confidence drops below your threshold - "grounding" mode kicks in and runs a web search to enrich the context before answering. Under the hood it's an open-weights LLM with a classifier fused on through post-training. That's the whole trick. Why it matters: 1. LLMs are built for chatbots, research and creativity - automation is a different sport. 2. A pipeline running 100K decisions a day needs speed and structured outputs, not eloquence. 3. Overconfidence is the silent killer of agent deployments - one hallucinated answer acted on with full certainty will burn you eventually. 4. An agent that says "signal too weak, human needed" is one you can actually put in production. Their thesis: Agent adoption is blocked by weak security, unreliability and painful setup. Sage attacks unreliability - exactly where mistakes cost money: guardrails, data pipelines, moderation, ops, fraud. Sage preview is already live - go test it yourself. The next wave of AI models won't win by knowing more! They'll win by knowing what they don't know.

@0xvikthor Let us know what you think and if you need some test creds

@levantolabs Big Chris cooked on this 🔥🔥🔥 checking it out

@withhimu @marco_derossi Thinking hurts the brain.

@levantolabs @marco_derossi Thinking hard. I built answerrank.so for boosting rankins on AI answers after generating over 10million views with organic SEO. Wondering where can it fit since there are many exact decisions AI is taking here.

@wallisi @marco_derossi @bigironchris Let us know if you'd like some creds!

@levantolabs Congrats @marco_derossi and @bigironchris 🔥

@0xvikthor @tsunamiii0 Monitoring

@tsunamiii0 @levantolabs Sure 💪

@marcusyul Escríbanos si tiene algún comentario.

UN MODELO DE IA QUE TARDA 200 MILISEGUNDOS EN DECIDIR ALGO. Y SABE DECIR "NO LO SÉ" se llama Sage, de Levanto Labs. le das hasta 32K tokens de contenido y una pregunta. responde en 200ms. 9 veces más rápido que un LLM tradicional. y siempre con un score de confianza pegado a la respuesta. si no está seguro, no inventa. lo dice. y escala la decisión a un humano. → sí/no, elección, ranking, tags o escala → con "grounding" activado, busca en la web si la confianza cae → por dentro: un LLM open-weights con un clasificador fusionado en el post-training ¿por qué te importa? porque automatizar procesos reales no necesita un chatbot más listo. necesita algo rápido, barato y que no se invente nada cuando no sabe. pensado para agentes, guardrails, pipelines de datos, moderación, riesgo y fraude. están contratando. preview ya disponible :)

@justinbebis Main character signals are everywhere

@levantolabs elite

@ExitLiqCapital that's true.

@levantolabs You do know tho

@frogmonkee the frogmonkee understands

@levantolabs This looks promising. I find that agent output routing is a hard problem. Lots of context gets dropped. Structured outputs could definitely change that.