Agent Office
Overview
The Agent Office is the primary workspace for interacting with your AI agents. It provides a full-page chat interface with a persistent conversation sidebar, real-time tool-call visibility, file attachments, voice-to-text input, and the ability to switch between agents mid-session.
Purpose
While the Global AI Assistant provides quick, contextual help from a slide-out panel, the Agent Office is designed for extended, focused work sessions with agents. It gives you:
- Full-page workspace — more room for reading responses and reviewing tool outputs
- Conversation persistence — all chats are saved and accessible from the sidebar
- Tool-call transparency — see exactly what actions the agent takes in real time
- File attachments — upload images and documents for the agent to analyze
- Voice input — speak your messages for hands-free operation
- Agent switching — change which agent you're talking to without losing context
How It Works
Interface Layout
Agent Office
├── Conversation Sidebar (left)
│ ├── New Chat Button
│ ├── Agent Selector (dropdown)
│ └── Conversation List
│ ├── Today
│ ├── Yesterday
│ └── Older
│ └── Each: title, timestamp, rename/delete actions
├── Main Chat Area (center)
│ ├── Agent Header (name, role, icon)
│ ├── Messages
│ │ ├── User Messages
│ │ ├── Agent Responses (with formatted content)
│ │ └── Tool Call Blocks (expandable)
│ └── Input Bar
│ ├── File Attachment Button
│ ├── Text Input
│ ├── Voice Button
│ └── Send Button
└── Welcome Screen (shown when no messages)
├── Agent Avatar & Bio
└── Suggested Prompts
Sending Messages
Type your message in the input bar and press Enter or click the Send button. The agent processes your message, optionally calls tools to perform actions, and streams its response in real time.
Input options:
| Method | How |
|---|---|
| Text | Type in the input bar; press Enter or click Send |
| Voice | Click the microphone button to record; transcription is automatic |
| File | Click the attachment button to upload images or documents |
Tool Call Display
When the agent uses tools (e.g., creating a dashboard, querying data, searching files), each tool call is displayed as an expandable block in the chat:
┌─ Tool Call: list_dashboards ──────────────────┐
│ Status: ✓ Completed │
│ ▶ Arguments: { } │
│ ▶ Result: Found 5 dashboards... │
└───────────────────────────────────────────────┘
Tool call states:
| State | Indicator | Description |
|---|---|---|
| Running | Spinning loader | Tool is currently executing |
| Completed | ✓ Green check | Tool finished successfully |
| Failed | ✗ Red X | Tool encountered an error |
You can expand any tool call to see its arguments and full result. This gives you complete visibility into what the agent is doing on your behalf.
Agent Loops
Agents can perform multi-turn tool-calling loops — calling multiple tools in sequence to complete complex tasks. For example, if you ask "Create a sales dashboard with a bar chart of monthly revenue from BigQuery":
- The agent calls a tool to explore your BigQuery tables
- It identifies the relevant dataset and columns
- It creates a new dashboard
- It adds a bar tile configured with the correct data source
- It returns a summary with a link to the new dashboard
Each step is visible as a tool call block in the chat, so you can follow the agent's reasoning in real time.
Conversation Management
The sidebar on the left provides full conversation management:
| Action | Description |
|---|---|
| New Chat | Start a fresh conversation with the current agent |
| Switch Conversation | Click any conversation in the sidebar to resume it |
| Rename | Click the pencil icon to rename a conversation |
| Delete | Click the trash icon to permanently delete a conversation |
Conversations are grouped by time: Today, Yesterday, and Older.
Agent Selector
Use the dropdown at the top of the sidebar to switch between agents. Each agent retains its own conversation history — switching agents shows that agent's past conversations.
File Attachments
Click the attachment button (📎) to upload files for the agent to analyze:
| Supported Types | What the Agent Can Do |
|---|---|
| Images (PNG, JPG, etc.) | Analyze content, extract text, describe visuals |
| Documents | Read and summarize content |
Attached files appear as thumbnails above the input bar before sending. You can remove an attachment before sending by clicking the ✗ button on the thumbnail.
Voice Input
Click the microphone button to start voice recording:
- Click mic — Recording begins (browser may request microphone permission)
- Speak your message — A recording indicator shows while capturing
- Click stop — Recording ends and audio is transcribed automatically
- Message is sent — The transcribed text is sent as your message
Maximum recording size is 25 MB.
Welcome Screen
When opening a new conversation, the Agent Office displays a welcome screen with:
- The agent's avatar, name, and bio
- Suggested prompts tailored to the agent's skills and capabilities
Click any suggested prompt to start the conversation with that topic.
Data Expectations
- Streaming responses — Agent responses appear word-by-word as they're generated
- Persistent history — All conversations are saved automatically; you can return to any past session
- Agent permissions — Agents operate with your permissions; they cannot access resources you don't have access to
- Credit usage — Each agent session and tool call consumes credits. See Agent Skills & Tools for detailed credit information
Use Cases
1. Dashboard Management
Ask your agent to "Show me all my dashboards" or "Create a new dashboard with sales metrics." Watch the tool calls in real time as the agent builds and configures tiles.
2. Data Exploration
Upload a spreadsheet image and ask "What trends do you see in this data?" or connect BigQuery and ask the agent to "Explore the sales dataset and summarize the schema."
3. Workflow Automation
Tell the agent to "Create a workflow that reads Gmail every hour and saves attachments to Google Drive." The agent will build the workflow node by node, connecting inputs to outputs.
4. Research with File Analysis
Attach a PDF or screenshot and ask the agent to extract key information, summarize findings, or compare data against your connected integrations.