Skip to main content

AI Agent Chat

An embedded AI assistant that answers questions about your workspace data and helps you operate the platform — browsing pages, filling forms, and submitting them with your explicit approval.

Overview​

The AI Agent Chat lives in a collapsible sidebar between the navigation menu and the page content, giving the platform a three-column layout:

┌────────┬──────────────┬──────────────────────┐
│  Menu  │   AI Chat    │        Canvas        │
│        │  (sidebar)   │   (platform pages)   │
└────────┴──────────────┴──────────────────────┘
AspectDetails
AvailabilityUsers with the AI_CHAT_ACCESS permission, when an AI provider is configured
ToggleRobot icon in the navigation sidebar (next to the logo)
PositionDockable to either side of the canvas
SizeDrag the divider to resize; maximize to full width; detach into a separate browser window
CollapsePress Escape to collapse the sidebar
StateOpen/collapsed state, position, and width are remembered across sessions
HistoryConversations are saved per user and can be reopened or deleted
note

If the chat icon does not appear in the header, either your role lacks the AI_CHAT_ACCESS permission or no AI provider has been configured for your tenant/server. See AI Chat Plugin for configuration.

What It Can Do​

Answer questions about your data​

Ask about tasks, projects, employees, time tracking, invoices, contacts, time off, and more:

  • "Which tasks assigned to me are overdue?"
  • "How many hours did the design team track last week?"
  • "Show me unpaid invoices for Acme Corp."

The agent queries the Gauzy API as you — it can only ever see data that your own user role and organization membership allow. Answers are always limited to what you could see yourself in the UI.

Open platform pages next to the chat​

The agent can navigate the canvas (the main content area) to any platform page while the conversation stays open — for example, "open the time tracking report" or "take me to the new invoice form".

Read and fill forms — you approve​

On the currently open page, the agent can:

  1. Read the form and its current values
  2. Fill fields based on your instructions
  3. Submit the form — but only after you click Approve in the chat
caution

Any action that modifies data — submitting a form, creating, updating, or deleting a record — always pauses and asks for your approval in the chat before it executes. Nothing is written on your behalf without an explicit click on Approve.

Attach files and discuss them​

Two ways to bring a document into the conversation:

  • Attach a file (paperclip icon) — the file is uploaded into the Documents hub (source: Chat), so it exists as a regular document from that moment, and it is attached to your next message.
  • Choose from the library (folder icon) — search and pick any document you already have in Documents.

Attached documents appear as chips on your message; the assistant reads them with its document tools, in your own permission scope, and can quote and cite them. A freshly uploaded file may need a moment of processing before its text is readable — the assistant will say so and you can simply ask again.

note

Attaching requires the Documents feature and, for uploads, the permission to create documents. Without Documents, a basic attach still works but the assistant can only see the file's name.

Voice dictation​

The microphone icon in the chat input records speech and transcribes it into the message box — a recording panel above the input shows the elapsed time and offers Cancel, Done, and an optional auto-send toggle. Dictation requires a configured provider with speech support (currently OpenAI).

Playground​

A dedicated playground page is available at /pages/playground for experimenting with the configured AI providers and models before (or alongside) using the embedded chat.

Permissions​

PermissionGrants
AI_CHAT_ACCESSUse the AI Agent Chat sidebar and playground
AI_CHAT_SETTINGSConfigure per-tenant AI provider API keys (Settings → AI Providers)

Permissions are assigned per role — see Custom Roles & Permissions.

Security Model​

  • The agent calls the Gauzy REST API with your own JWT — role-based access control and tenant isolation always apply.
  • Read-only questions run directly; mutating tools require in-chat approval.
  • Provider API keys are stored encrypted on the server and are never exposed to the browser.