> ## Documentation Index
> Fetch the complete documentation index at: https://docs.nexalytica.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Chat Composer: Agents, Tools & Deep Thinking Modes
> Control how the AI answers your questions. Choose agents, enable Deep Thinking, toggle web search, and manage tools from the chat composer.
The composer is the bar pinned to the bottom of every chat. It is where you write messages and configure exactly how the AI agent approaches your question — from the reasoning depth it applies, to which tools it can use, to which files it can read. All composer settings are remembered per chat, so each conversation keeps its own configuration without affecting other threads.
## Composer layout
Active settings appear as removable pills (chips) above the message box. You can see at a glance which thinking level, agent, and tools are enabled for the next message.
### Message box
| Interaction | What it does |
| ----------------------------- | ---------------------------------------------------- |
| Type in the box | Compose your question or instruction |
| **Enter** | Sends the message |
| **Shift + Enter** | Inserts a new line without sending |
| **Send button** (arrow icon) | Sends the message |
| **Stop button** (square icon) | Cancels a response that is currently being generated |
The **Stop** button is only visible while the agent is generating a reply. It disappears once the response is complete.
***
## Deep Thinking modes
The **Deep Thinking** button (brain icon) cycles through three levels of reasoning depth. Click it repeatedly to advance through the cycle, or remove the active chip to return to the default.
**Fast conversational mode.**
The agent responds quickly and can still use tools. Choose this level for straightforward questions, quick lookups, and iterative back-and-forth conversations where speed matters.
**Parallel multi-branch reasoning.**
The agent breaks your question into branches and works on them in parallel before synthesising an answer. This level is well suited to multi-part questions and analytical requests that benefit from broader exploration.
**Sub-agent delegation — the most thorough mode.**
The agent hands distinct parts of your question to specialised sub-agents, each focused on a specific aspect of the problem. Results are aggregated into a comprehensive answer. Use this for complex investigations where depth matters more than speed. Expect longer response times.
Start with **Deep Thinking** for analytical questions. Escalate to **2X Deep Thinking** only when you need exhaustive, multi-angle coverage and can wait for a slower response.
***
## Choosing an agent
Click the **Agent picker** button (robot icon) in the composer to open the agent list. The list shows every AI agent approved for your department by your organisation's admins.
* Select an agent to make it active for subsequent messages in this chat.
* The active agent's name appears both as a chip above the message box and in the conversation header.
* Different agents may have access to different tools, data integrations, and reasoning capabilities.
Only agents approved by your organisation's admins appear in the picker. Contact your admin if you need access to an agent that is not listed.
***
## Tools panel
Click the **Tools** button (wrench icon) in the composer to view and manage the tools available for the next message. The button label shows **X/Y**, where **X** is the number of enabled tools and **Y** is the total available.
| Control | What it does |
| ----------------- | ---------------------------------------------------------------- |
| **Tool checkbox** | Untick to disable a specific tool for this message |
| **Reset** | Returns all tool selections to the agent's default configuration |
The tools list is determined by the agent you have selected and by which tools your organisation's admins have approved. You cannot enable a tool that has not been approved.
Some data-focused tools only become available after you attach at least one data source to the chat. Open **Sources (N)** in the conversation header to attach sources.
***
## Web Search
Toggle **Web Search** in the composer to let the agent search the web in addition to your connected data sources. When enabled, the agent can retrieve current information from outside your organisation's data to supplement its answers.
* Web Search appears as an active chip above the message box when it is on.
* Click the **×** on the chip, or toggle the button again, to turn it off.
***
## Attaching files and data sources
Click the **paperclip** icon in the composer to open the Attachments panel. Upload a PDF, Excel, CSV, or text file (up to 100 MB each) to make it available to the agent for the current conversation. Attached files are also visible in the **Artifacts** panel in the conversation header.
Click **Sources (N)** in the conversation header to open the Sources panel. Tick the data sources you want the agent to read. The count **N** updates to show how many sources are currently active.
***
## Active chips
Every enabled composer setting appears as a chip above the message box. Chips give you a live summary of how the next message will be processed.
| Chip | What it represents |
| ------------------- | ----------------------------------------------- |
| Thinking level chip | Current Deep Thinking mode (hidden when off) |
| Web Search chip | Web search is enabled |
| Agent chip | The currently selected agent |
| Tool chips | Any tools that differ from the agent's defaults |
Click the **×** on any chip to remove that setting and revert it to its default.
***
## Permissions
* Type and send messages
* Stop a reply in progress
* Switch Deep Thinking levels
* Toggle Web Search
* Pick an agent
* Adjust tool selections
* Attach files and data sources
Viewers have read-only access to the chat. They **cannot** send messages, stop replies, or change any composer settings.
Admins control which agents and tools are available organisation-wide. Only agents and tools they have approved appear in the composer for any user.