Agents in Vibe

Updated November 2, 2026 — Le Chat is now Vibe. As of May 28, 2026, Le Chat is now Vibe (still at chat.mistral.ai): a single agent for work and code, carrying over your conversations, settings, and subscriptions. The legacy Le Chat experience remains accessible in Chat mode; in Work mode, agents are now skills (invoked with @), memories are the knowledge base, and deep search is the Deep Research skill. Plans are now called Mistral Free, Pro, Team, and Enterprise. This lesson now uses the name Vibe; some menus described here may have changed names or locations.

Create specialized assistants

In a standard conversation, you start from scratch: you have to re-explain the expected role, format, and constraints. An agent cuts out this repetition. It’s a preconfigured assistant, powered by custom instructions and specific tools, built for a given task—either yours or one shared by another user in the community.

Two components make up an agent. Instructions define its behavior, tone, and expertise: “You are an expert in French corporate law. You respond precisely, citing relevant articles of law” creates a very different assistant than “You are an SEO writing assistant. All content you produce is optimized for search engines.” Tools extend its capabilities beyond text generation.

Build your first agent

Creation follows a short sequence:

1. Go to the “Agents” section in Vibe, then click “Create an agent”
2. Give it a name and a clear description
3. Write the instructions that define its behavior, then select the tools to enable
4. Upload reference documents if needed, then test your agent with a few sample queries

Writing the instructions makes all the difference between a useful agent and a gimmick. A good instruction first establishes the role without ambiguity: “You are an assistant specialized in writing marketing content for the B2B SaaS sector.” It describes the expected output format, which saves you editing every response: “Every response must include a catchy title, three paragraphs of 100 words maximum, and a call to action.” It sets explicit constraints, because what you forbid matters as much as what you request: “Never use superlatives. Prefer hard data to general statements. Always cite your sources.” Finally, it sets the tone: “Professional but warm tone, plain language, no unnecessary jargon.”

Don’t skip the testing phase. You judge an agent on varied queries, including edge cases: that’s where you’ll see that an instruction that’s too vague allows off-topic responses, or that a poorly phrased constraint blocks legitimate requests.

Agents that deliver

An industry monitoring agent relies on Web Search and a time-bound instruction: “You monitor news in the [X] sector. Every morning, you provide a summary of the five most important stories from the monitoring period with sources.” Provide it with the list of competitors and sector keywords, and you replace half an hour of scattered reading with a sourced summary.

A data analyst agent uses Code Interpreter, with an instruction that specifies both the deliverable and the audience: “You analyze the data files provided to you. You produce clear visualizations and actionable insights. You explain your results to a non-technical audience.” The data dictionary and your benchmark KPIs will make it speak the language of your dashboards. A visual creator agent, on the other hand, combines Image Generation and a strict style guide—“You create marketing visuals compliant with our brand guidelines. Colors: navy blue and gold. Style: minimalist and premium.”—which you include along with a few already approved visuals.

A legal agent finally shows the importance of defining what the agent is not: “You assist in drafting and analyzing contracts. You flag problematic clauses and propose alternatives. You are not a lawyer and you always recommend validation by a legal expert.” Powered by Document Library with your contract templates, standard clauses, and relevant statutes, it speeds up review without replacing legal counsel.

Once your agent is proven, share it with your team or the community, either as is or for others to duplicate and adapt. Conversely, the community gallery is worth browsing before you start from scratch.

Key takeaways

• Agents are specialized assistants with custom tools and instructions
• Four main tools: Web Search, Code Interpreter, Image Generation, Document Library
• Precise and structured instructions are key to an effective agent
• Test your agent with varied queries before deploying it
• Share your successful agents and explore those from the community