🆕 Le Chonk brings the brains. Mistral Studio gets it working. Introducing managed agents and apps, plus new capabilities for workflows, enterprise connections, observability and evaluations in Studio. Build faster. Deploy where you choose. Keep control. See what’s possible. 🧵

It's cool, but have you anything planed for Vibe Work or will we have to keep using GLM 🥲

Managed agents plus built-in evals is an amazing stack.

Managed agents, apps, workflows, observability, evaluations. Every AI lab is now one press release away from being Salesforce. At least Le Chonk has a better name haha

TF Is happening inside Mistral, banger after banger in this period

Le Chonk getting presented to the enterprise

managed agents in studio is a big one. real q tho, can you cap spend per agent or does le chonk get the company card and go wild

Le Chonk is doing half the marketing already.

The actual enterprise harness

O gancho pra produção: um código, três modos (local, self-hosted com agents serve, managed com agents deploy). Sessão aguenta horas/dias, retoma do último passo após crash, e aprovação humana espera com o app fechado. O produto aqui é a infra de agente longo, não só o modelo.

Can the agent runtime run fully on-prem?

Les agents gérés avec observabilité et évaluations, c’est ce qui manque souvent en entreprise : savoir ce que l’agent a fait et le mesurer. Vous proposez un hébergement chez le client, ou tout reste chez Mistral ?

Managed agents with evals and observability in one place is the part that caught my eye. Can you plug in a non Mistral model for the agents, or is it Mistral only?

Powerful combination! Smarter agents, seamless deployment, and more control make this a big step forward for AI development.

le chonk was the model. studio is the missing piece: managed agents, apps, workflows, and you can keep it in your own environment. brains without a place to run them was the gap

agents you can host wherever you want is a big deal for indian companies scared of data leaving the country tbh

deploy where the data lives

Congrats on the launch! For keep control, give every managed agent's enterprise connections a named owner and a fresh review whenever its access grows.

One default worth checking before long runs: Mistral's docs say deployed agent sessions end after 24 hours, or 4 hours idle. Both limits can be raised.

Deploy where you choose is the part that matters for real companies. Managed agents only work in production if every step can be seen, traced and undone when the agent gets it wrong.

managed agents sound easy until one goes off the rails and nobody can see why. observability and evals shipping in the same launch is the part worth watching

Managed agents live or die on egress and per-token observability billing. 'Deploy where you choose' only holds if enterprise connections don't route inference back through your cloud.

on est de retour ?

observability and evals built into studio is the adult bit. most agent demos skip that and then wonder why nobody puts them in production

deploy where you choose is the line I would test first, managed agents usually means the runtime lives on their side even when the weights dont

Managed agents live or die on observability and evals, not the model. Putting those inside Studio is the right priority for enterprises that must audit what an agent actually did.