Mistral Studio Adds Managed Agents, Workflows and Apps

Mistral AI, the Paris-based lab behind the Le Chonk model, just turned its enterprise platform into a full agent workshop. On November 5, 2026, Mistral AI expanded Mistral Studio with managed agents, a natural-language workflow builder and one-command app deployment, alongside new tools for enterprise data connections, observability and evaluations. The pitch is simple: one place to build an agent, wire it into company systems, and run it where your data already lives: in Mistral's cloud, your own VPC, or on-premises.

The release lands a month after Studio itself debuted in October 2025, and it arrives with Mistral Large 4, the lab's trillion-parameter flagship nicknamed Le Chonk, as the default brain. The question for teams is not whether the feature list is long. It is what is actually usable today, what is still preview, and what the bill looks like. This article works through all three, using Mistral's own announcement and documentation, checked on November 7, 2026.

What Mistral announced on November 5

Mistral's announcement, titled "Le Chonk brings the brains. Studio puts it to work," frames Studio as one platform with three building blocks: agents that investigate and act, workflows that coordinate execution and approvals, and apps that give employees a way to use those systems. The documentation release notes list eleven new entries dated November 5, 2026 (ten in public preview and one, HTTP API connectors, in private preview), plus a generally available rendering of Studio apps inside Vibe Code Web, dated October 14.

The headline items are managed agents, the workflow builder and the Mistral Apps CLI. Around them sit the quieter additions that matter for production use: a managed search index, an Evaluation SDK with offline and online evaluations, managed datasets, tracing for managed agents, webhooks and event triggers, and a Secrets Manager. Mistral says customers and partners including ABANCA, Accenture, Alten, BNP Paribas, Capgemini, CMA CGM and Tata Consultancy Services already use Studio in production.

That customer list is Mistral's own statement, not an independent audit, and the same applies to the two time-saving examples in the announcement. Still, the direction is clear. Mistral is selling Studio as the place where a pilot becomes a system a company can operate, measure and keep under its own control.

Managed agents: write once, run three ways

The centerpiece is Managed Agents, built on a new Python Agents SDK, mistralai-agents. You define the model, instructions, tools and environment once, then run the same code in three modes: inside your own process while developing, self-hosted with agents serve, or Mistral-managed with agents deploy. The agents init command scaffolds a working project.

Once deployed, execution is durable and runs in the background. An agent can work for hours or days, resume from its last finished step after a crash, and wait for a person's approval before continuing, then pick up later. Mistral stores the session, so you host no database. Agents can be equipped with skills that capture a team's procedures, sandboxes where they can run code, subagents to delegate parts of a task, and hooks that add custom logic at key points in execution. The runtime is the same infrastructure Mistral says supports millions of Mistral Vibe sessions each week. That figure is company-reported.

Approvals are the feature to notice first. You choose which tool calls need approval from code or from a person, and you can steer or interrupt a turn while it runs. For an agent with access to company systems, that control is the difference between a demo and something a compliance team will sign off on.

Mistral's own engineering team built an incident-investigation agent in Studio as the worked example. When an alert fires, the agent follows the team's runbook, asks for approval before taking sensitive actions, and posts its findings in Slack. Mistral says it saves the team hundreds of hours each month. That is a company-reported number with no independent verification, but it paints a useful picture of the intended use: long, multi-step work that waits on people at the risky points.

Workflows: describe the process in plain language

Workflows are the deterministic counterpart to agents. A workflow defines which steps follow fixed rules, where an agent reasons, and when a person decides. Interrupted runs resume where they left off, and a visual graph shows each workflow's logic.

The new workflow builder opens authoring to people who do not write code. You describe a process in natural language, and Mistral Vibe writes the workflow, tests it and deploys it, with code you can inspect and edit at every step. The release also adds event triggers, so workflows and agents can start when something happens in a business system, and webhooks defined from the Studio UI or the API. Built-in ingestion covers services such as Notion and Lucca, with identity resolution, message normalization and deduplication handled for you.

The worked example here is a product-operations workflow that sorts ideas from internal teams and customer feedback. A model identifies bugs and feature requests, then routing rules send each submission to the right team. Mistral says this saves its product operations team about 35 hours a week. That is the company's own figure, not an independent measurement.

If you already build automations in tools such as Zapier or n8n, the difference is the target audience. Those tools automate personal and small-team tasks; Mistral's workflows are built for enterprise processes that touch internal systems, need approval steps, and must keep running when something fails.

Apps: one command to an enterprise-ready link

The third block is apps. Coding agents make it easy to build an application, but putting it to work across an enterprise means meeting requirements for security, access and reliability. Studio provides hosting, authentication and access controls, and teams build with pre-approved, reusable components, which reduces the amount of new code that needs security review.

The new Mistral Apps CLI scaffolds an app from a capability registry, runs it locally, and deploys it to Studio in one command. Apps can use agents and workflows, retrieve enterprise documents through search indexes, and access shared context in Studio. Developers test locally, then deploy to make the application available at its own link or inside Mistral Vibe. Studio apps also now render inside Vibe Code Web, where a live workflow graph sits beside the builder conversation. That rendering is the one generally available item in this release.

Mistral's example is a staffing app built by its Applied AI team. It uses an agent to recommend people for a project and explain how their experience fits, giving staffing leads a shortlist with supporting evidence. The honest read is that this is the least surprising part of the release, and the most useful. Internal tools are where most enterprise AI value sits, and a one-command deploy with auth included is the part teams actually feel.

The observability layer: tracing, evaluations and enterprise context

The features that decide whether Studio survives contact with production sit around the edges. Managed agents now send agent, model and tool spans to Studio by default, with no extra package or configuration. Each turn becomes a trace, and spans are redacted on the worker before they leave. Mistral says default redaction masks common secrets and personal data, with custom policies available.

Evaluations arrive in two forms. Offline evaluations run from the Evaluation SDK or a workflow eval plugin, using datasets you build from CSV or JSONL files, free-format records, or production telemetry. You can turn a trace span into a dataset record in one click. Online evaluations score real sessions in production, so you can catch outcome regressions before customers report them. Mistral's documentation describes traces and evaluations as available to Enterprise-tier organizations, so this layer may not be open to every plan.

Enterprise context gets three additions: HTTP API connectors (private preview), which let you register a web API by describing its operations instead of building an MCP server; a managed search index, a hosted vector index provisioned from a name, an embedding model and an optional schema; and Secrets Manager, which stores and injects API keys and credentials into managed deployments with role-based access control, versioning and audit logs.

What it costs

There is no bundled Studio price, and Mistral has published no per-agent or per-run fee for the new features. What the documentation does state is the access rule: during public preview, managed deployments require a paid plan (pay-as-you-go or Enterprise) and are subject to a quota. Coverage of the launch also notes that apps cannot be created on Free, Pro or Team plans without pay-as-you-go billing enabled. Treat the preview as free to try in Studio, and billable the moment you deploy something managed.

The costs you can pin down are the subscription tiers and the model rates. Mistral's pricing page lists a Free plan, Pro at $14.99 a month, Team at $24.99 per user per month, and Enterprise as custom pricing. The Free plan includes access to Studio, according to the pricing page's FAQ. The paid plans also bundle monthly API credits.

Underneath, the meter runs on tokens. Mistral Large 4, the flagship model Studio is built around, costs $0.70 per million input tokens, $0.07 per million cached input tokens and $2.10 per million output tokens on Mistral's API, a rate card verified on Mistral's own API pricing page on November 7, 2026. That is roughly half the $1.36 input and $4.18 output list price the same page showed before October 29, when Mistral cut the rate alongside the open-weights release.

GBP and CAD figures are approximate conversions at November 7, 2026 market rates (1 USD = 0.7471 GBP; 1 USD = 1.4117 CAD). Mistral bills in USD, so the sterling and Canadian figures will move with the exchange rate. The token bill is the one to model before you commit: an agent that runs for hours, calls tools, and re-reads context will burn far more input tokens than a chatbot, and cached input at a tenth of the price is the lever that keeps it affordable.

What is still in preview, and what that means for you

Almost everything in this release is in public preview, and Mistral's own definition matters: the feature can be used now, but its API, behavior and limits may change, and an update may require code changes on your side. HTTP API connectors are a step earlier, in private preview by invitation. Only the rendering of Studio apps inside Vibe Code Web is generally available.

That status has practical consequences. Do not build a business-critical process on a public-preview agent without a fallback, and expect the Agents SDK to change. The older Agents API guides are now filed under a Deprecated section in Mistral's documentation, so teams already using the earlier API should plan a migration rather than building new work on it.

Deployment is the other caveat. The announcement promises deployment in Mistral Cloud, your own VPC, on-premises, or a mix. That is the core of Mistral's sovereignty pitch. But coverage of the launch notes that the new managed features reach private cloud and on-premises deployments at the end of November, with Mistral Cloud first. If your data cannot leave your perimeter, confirm the on-premises date for the specific feature you need before you plan around it.

Who should try it now, and who should wait

Try it now if your team already runs on Mistral's models and you have a process that is long, repetitive, and currently eats a person's week: incident triage, feedback routing, document review, or internal staffing. The workflow builder is the fastest way in: describe the process, inspect the generated code, and deploy it to a managed deployment. The free Studio access on the Free plan makes that experiment cheap, as long as you stay off managed deployments until you understand the token bill.

Wait if you need a stable API contract, a private-cloud or on-premises deployment this month, or the tracing and evaluation tooling on a non-Enterprise plan. The preview labels are not decoration, and the on-premises rollout is still weeks away. And if your agents need to touch production systems, set the approval rules before the first run. The human-in-the-loop controls are the whole point of this release, and they only protect you if you turn them on.

One more thing to check before you start: the model. Mistral Large 4 is the flagship, but it is still labeled public preview on its model card, and its weights ship under the Mistral Ecosystem Open-Weight License, which sends businesses operating a model-as-a-service offering to a separate commercial license. For more on the model itself, see our Mistral Large 4 review. If your agent needs voice, Mistral's Voxtral 3 handles transcription from $0.0035 a minute. For a free, open alternative to the workflow layer, Google's Data Agent Kit offers MCP tools for data-cloud services at no platform cost, though it is a toolkit, not a managed runtime.

Frequently asked questions

What is Mistral Studio?

Mistral Studio is Mistral AI's enterprise platform for building and deploying agents, workflows and apps connected to company data and systems, with deployment in Mistral Cloud, your own VPC, on-premises, or a mix. It launched in October 2025 and expanded with managed agents and apps on November 5, 2026.

Is Mistral Studio free?

Studio access is included on Mistral's Free plan, and you can build agents and workflows there. Managed deployments, which run your agents and workflows on Mistral's infrastructure, require a pay-as-you-go or Enterprise plan during public preview, and usage is billed on tokens.

What does Mistral Large 4 cost in Studio?

The API rate is $0.70 per million input tokens, $0.07 per million cached input tokens and $2.10 per million output tokens, verified on Mistral's API pricing page on November 7, 2026. That is about half the $1.36 and $4.18 list price shown before October 29.

Can I deploy Studio agents on-premises?

That is the plan, and the announcement promises cloud, VPC, on-premises or mixed deployments. The new managed features run on Mistral Cloud first, with private cloud and on-premises availability reported for the end of November 2026. Confirm the date for your specific feature before planning around it.

How is Mistral Studio different from the Agents API?

The Agents API is a set of endpoints for building agents in code. Studio adds a managed runtime, durable background execution, approvals, tracing, evaluations and one-command app deployment on top, plus a visual workflow builder for non-developers. Mistral's docs now file the older Agents API guides under a Deprecated section.

Which model powers Mistral Studio?

Mistral Large 4, nicknamed Le Chonk, is the flagship model for coding, defensive cybersecurity and reasoning over complex documents and business tasks in Studio. It is a 1.05-trillion-parameter mixture-of-experts model with 52 billion active parameters and a 1-million-token context window.

Mistral Studio's November 5 release is not a new model. It is the scaffolding that lets enterprises actually use the model they already have. Managed agents, plain-language workflows and one-command apps are all useful, and the approval and tracing controls show Mistral understands what blocks enterprise adoption. The price of admission is patience with preview labels and a careful eye on the token meter. Start with one process, measure it, and let the workflow builder earn the next one. Read Mistral's announcement and the full release notes before you build.

Sources

• Mistral AI: Le Chonk brings the brains. Studio puts it to work. (November 5, 2026)
• Mistral Docs: Release notes (November 5, 2026 entries)
• Mistral: Pricing (Free, Pro $14.99, Team $24.99, Enterprise)
• Mistral Docs: API pricing (Mistral Large 4: $0.70 input, $0.07 cached, $2.10 output per 1M tokens)
• RateMorph: USD to GBP, November 7, 2026 (1 USD = 0.7471 GBP)
• SonicWallet: USD to CAD, November 7, 2026 (1 USD = 1.4117 CAD)
• DataStudios: Mistral Studio adds managed agents, workflow builder and apps in public preview
• DEV Community: Mistral Studio adds managed agents, workflows and apps for enterprise automation
• Injoys: Mistral AI Studio Adds Agents in Public Preview