Mistral Le Chat Work Mode Guide — Multi-Step AI Tasks with Tool Integration (2026)
Le Chat Work Mode is Mistral’s agentic interface for multi-step tasks that span multiple tools and data sources. Instead of answering a single question, Work Mode breaks complex requests into subtasks, calls tools in parallel, synthesizes results, and presents a structured output — all while showing you exactly what it is doing and why.
It runs on Mistral Medium 3.5 as the execution backend. The model orchestrates tool calls, manages state across steps, and decides when to ask for human approval before taking sensitive actions. Think of it as an AI assistant that can actually do things across your work tools, not just talk about them.
This guide covers how Work Mode works, what you can do with it, and how it compares to similar features from OpenAI and Anthropic.
For details on the underlying model, see the Mistral Medium 3.5 complete guide.
What is Work Mode
Work Mode is a toggle in Le Chat (Mistral’s consumer and business chat interface) that switches the model from single-turn Q&A to multi-step agentic execution. When enabled, the model can:
- Call multiple tools in a single turn
- Execute tool calls in parallel when there are no dependencies
- Chain results from one tool into the next
- Access connected services (email, calendar, Jira, Slack, Google Workspace, etc.)
- Browse the web and synthesize information from multiple sources
- Generate files, documents, and structured outputs
- Draft actions and wait for your approval before executing sensitive operations
The key difference from standard chat: Work Mode does not just generate text. It takes actions. It reads your email, checks your calendar, searches Jira, drafts a response, and creates a ticket — all in one turn if you ask it to.
How it works under the hood
Execution architecture
When you submit a request in Work Mode, the flow is:
- Planning: Medium 3.5 analyzes your request and breaks it into subtasks
- Tool selection: The model identifies which tools are needed for each subtask
- Parallel execution: Independent subtasks run simultaneously (e.g., fetching email and checking calendar at the same time)
- Sequential chaining: Dependent subtasks wait for prerequisites (e.g., creating a Jira ticket based on email content)
- Synthesis: Results from all subtasks are combined into a coherent response
- Approval gates: Sensitive actions (sending emails, creating tickets, modifying data) pause for your confirmation
Visibility and transparency
Every step is visible in the Le Chat interface. You see:
- Thinking rationale: Why the model chose specific tools and approaches
- Tool calls: Exactly which APIs were called, with what parameters
- Intermediate results: What each tool returned
- Decision points: Where the model made choices about how to proceed
This transparency is not optional — it is always on. You can inspect every tool call and understand exactly what happened. This matters for enterprise adoption where audit trails are required.
Connectors
Work Mode uses connectors to integrate with external services. Connectors are enabled by default for your connected accounts. Currently supported:
| Connector | Capabilities |
|---|---|
| Gmail / Outlook | Read, search, draft, send emails |
| Google Calendar / Outlook Calendar | Read events, create events, check availability |
| Slack | Read channels, post messages, search history |
| Jira | Create issues, update status, search tickets, add comments |
| Google Drive / OneDrive | Read documents, search files, create documents |
| Notion | Read pages, create pages, search workspace |
| GitHub | Read repos, create issues, search code |
| Web search | Search the internet, fetch and parse web pages |
| Code execution | Run Python code in a sandboxed environment |
You manage connector permissions in Le Chat settings. Each connector can be individually enabled or disabled. When a connector is active, Work Mode can use it without asking permission for read operations. Write operations (sending emails, creating tickets) require your approval.
Use cases
Cross-tool workflows
The most powerful use case is workflows that span multiple tools. Examples:
Morning briefing:
“Summarize my unread emails, list today’s meetings, and flag any Jira tickets assigned to me that are overdue.”
Work Mode calls Gmail, Google Calendar, and Jira in parallel, then synthesizes a single briefing. This replaces the manual process of checking three different apps every morning.
Meeting preparation:
“I have a meeting with Acme Corp at 2pm. Pull the latest emails from their team, check our Jira board for open issues related to their project, and draft an agenda.”
The model reads your calendar to confirm the meeting, searches email for recent Acme Corp correspondence, queries Jira for relevant tickets, and generates a structured agenda document.
Project status report:
“Generate a weekly status report for Project Atlas. Include completed Jira tickets, open blockers, and any relevant Slack discussions from #project-atlas this week.”
Work Mode queries Jira for ticket status changes, searches Slack for the channel history, and produces a formatted report.
Research and synthesis
Work Mode excels at gathering information from multiple sources and producing structured output.
Competitive analysis:
“Research the latest pricing changes from Anthropic, OpenAI, and Google for their flagship models. Compare with our current Mistral costs and summarize in a table.”
The model searches the web for current pricing pages, extracts the relevant data, and produces a comparison table. It can also pull your usage data from connected tools to make the comparison specific to your workload.
Technical evaluation:
“Find the top 5 open-source vector databases, compare their performance benchmarks, licensing, and cloud hosting options. Output as a structured brief I can share with the team.”
Web search, multiple page fetches, data extraction, and synthesis — all in one turn. The output is a document you can copy into Google Docs or Notion.
Inbox triage
This is where Work Mode saves the most time for individual users.
Email triage:
“Go through my unread emails. For each one: if it’s a support request, create a Jira ticket. If it needs a reply, draft one. If it’s informational, summarize it in one line. Give me the results as a table.”
Work Mode reads each email, classifies it, takes the appropriate action (drafting replies, creating tickets), and presents a summary table. You review the drafts and approve or edit before anything is sent.
Slack catch-up:
“Summarize what happened in #engineering and #product channels today. Flag anything that needs my attention.”
The model reads channel history, identifies important messages, and produces a prioritized summary. This is particularly valuable after a day off or when you are in too many channels to read everything.
Content creation with research
Blog post drafting:
“Write a technical comparison of PostgreSQL vs CockroachDB for our engineering blog. Research current benchmarks, include code examples for common operations, and cite sources.”
Work Mode searches for recent benchmarks, reads documentation, generates code examples (and can run them in the sandbox to verify), and produces a draft with citations.
Documentation updates:
“Check our API docs in Google Drive against the current codebase on GitHub. Flag any endpoints that are documented but no longer exist, or exist but are not documented.”
Cross-referencing Google Drive documents with GitHub repository contents — a task that would take hours manually.
Approval workflow for sensitive actions
Work Mode distinguishes between read and write operations. Read operations (fetching emails, searching Jira, browsing the web) execute automatically. Write operations pause for your approval.
When the model wants to take a sensitive action, you see:
- The exact action it wants to take (e.g., “Send email to john@acme.com”)
- The content of the action (e.g., the email draft)
- An approve/reject button
You can edit the content before approving. This prevents the model from sending an email with incorrect information or creating a Jira ticket with the wrong priority.
The approval threshold is configurable in enterprise plans. Admins can set policies like “always require approval for external emails” or “auto-approve Jira ticket creation within the team project.”
For more on how AI agents handle approval workflows, see what is an AI agent.
Comparison with alternatives
Work Mode vs ChatGPT plugins/GPTs
ChatGPT’s plugin ecosystem and custom GPTs offer similar tool integration, but with key differences:
| Feature | Le Chat Work Mode | ChatGPT plugins/GPTs |
|---|---|---|
| Parallel tool calls | Yes, native | Limited |
| Thinking visibility | Full rationale shown | Partial |
| Enterprise connectors | Jira, Slack, Google Workspace | Varies by plugin |
| Approval workflow | Built-in for write actions | Plugin-dependent |
| Data residency | EU (Mistral servers) | US (OpenAI servers) |
| Underlying model | Mistral Medium 3.5 | GPT-4o/GPT-4.5 |
| Custom connectors | Enterprise plan | Custom GPTs |
Work Mode’s main advantage is transparency and EU data residency. ChatGPT’s advantage is a larger plugin ecosystem.
Work Mode vs Claude artifacts
Claude’s artifacts are focused on content creation — generating documents, code, and visualizations within the chat interface. They do not integrate with external tools. Work Mode is fundamentally different: it is about taking actions across your work tools, not just generating content.
The two are complementary rather than competitive. Use Claude artifacts for content creation. Use Work Mode for cross-tool workflows.
Work Mode vs custom AI agents
Building custom agents with LangChain, CrewAI, or similar frameworks gives you more control but requires engineering effort. Work Mode is a managed solution — no code required, connectors pre-built, approval workflows included.
For teams that need custom logic or proprietary tool integrations, building your own agent is still the right choice. For standard business workflows across common tools, Work Mode is faster to deploy.
For guidance on the build-vs-buy decision, see AI agent build vs buy.
Pricing
Work Mode is available across Le Chat plans:
| Plan | Price | Work Mode access | Limits |
|---|---|---|---|
| Free | $0/month | Basic (limited tool calls) | 50 Work Mode requests/day |
| Pro | $14.99/month | Full | 500 Work Mode requests/day |
| Team | $25/month per seat | Full + shared connectors | 1,000 requests/day per seat |
| Enterprise | Custom | Full + custom connectors + admin controls | Custom limits |
Work Mode requests consume more compute than standard chat messages because they involve multiple model calls and tool executions. The daily limits reflect this higher cost.
Enterprise plans include custom connector development, SSO integration, admin dashboards for monitoring usage, and configurable approval policies.
Getting started
Step 1: Enable Work Mode
Open Le Chat (chat.mistral.ai). Click the mode selector in the chat input area and switch from “Chat” to “Work.” The interface changes to show a tool panel on the right side.
Step 2: Connect your tools
Go to Settings → Connectors. Connect the services you want Work Mode to access. Each connector uses OAuth — you authorize Le Chat to access your account, and permissions are scoped to what the connector needs.
Start with email and calendar. These are the most immediately useful connectors and have the lowest risk (read operations are the primary use case).
Step 3: Start with simple requests
Begin with read-only workflows to build trust in the system:
- “Summarize my unread emails”
- “What meetings do I have today?”
- “Search Jira for tickets assigned to me”
Once you are comfortable with how Work Mode reads and synthesizes information, move to write operations:
- “Draft a reply to the email from Sarah about the Q2 budget”
- “Create a Jira ticket for the bug reported in #engineering”
Step 4: Build complex workflows
Combine multiple tools in a single request:
- “Check my email and calendar for conflicts this week, then draft a Slack message to the team about my availability”
- “Research competitor pricing, create a comparison document in Google Drive, and share it with the product team”
Tips for effective Work Mode usage
- Be specific about output format. “Give me a table” or “Output as bullet points” helps the model structure its response.
- Name the tools you want used. “Check Jira and Slack” is better than “look into the project status” — it reduces ambiguity.
- Review before approving. Always read drafts before approving send/create actions. The model is good but not perfect.
- Use it for triage, not final decisions. Work Mode is excellent at gathering and organizing information. Use it to prepare for decisions, not to make them.
For more on how Mistral’s agentic capabilities extend to development workflows, see the Mistral Vibe 2 remote agents guide.
FAQ
Does Work Mode have access to all my email and files?
Only what you authorize. Each connector uses OAuth with scoped permissions. You can revoke access at any time in Settings → Connectors. Read operations happen automatically when you make a request; write operations require your explicit approval. Mistral’s privacy policy states that connector data is not used for model training.
Can I use Work Mode with on-premise tools?
On the Free and Pro plans, only cloud-based connectors are supported. Enterprise plans support custom connectors that can integrate with on-premise tools through secure tunnels or API gateways. Contact Mistral sales for on-premise connector development.
How does Work Mode handle sensitive data in emails and documents?
Data accessed through connectors is processed by Mistral Medium 3.5 on Mistral’s EU infrastructure. It is subject to Mistral’s data processing agreement and GDPR compliance. Data is not stored beyond the session unless you explicitly save outputs. For highly sensitive data, consider the Enterprise plan with dedicated infrastructure or self-hosting the model.
Is Work Mode available on mobile?
Yes, Le Chat has iOS and Android apps with Work Mode support. The mobile interface shows the same tool calls and approval workflows as the desktop version. However, complex multi-tool workflows are easier to review and manage on a larger screen.
Can I automate Work Mode workflows to run on a schedule?
Not directly in Le Chat. Work Mode is interactive — it requires you to initiate requests and approve actions. For scheduled automation, use Mistral’s API with Medium 3.5 and build your own orchestration layer. The API supports the same tool-calling capabilities that power Work Mode, but you control the execution loop.
How does Work Mode compare to Microsoft Copilot?
Microsoft Copilot is deeply integrated with Microsoft 365 (Outlook, Teams, Word, Excel). If your organization is fully on Microsoft 365, Copilot has tighter integration with those specific tools. Work Mode is tool-agnostic — it connects to Google Workspace, Slack, Jira, and other non-Microsoft tools equally well. Work Mode also offers full transparency into its reasoning and tool calls, which Copilot does not always provide. The choice depends on your tool ecosystem.