> ## Documentation Index
> Fetch the complete documentation index at: https://docs.quadrillion.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Agents

> Run parallel experiments and delegate work to independent agents.

Qualia's agent system lets you go beyond single-threaded conversations. You can run multiple experiments in parallel, delegate sub-tasks to independent agents, and scale your research.

## What is an agent?

An agent is a chat session that can work independently. Each agent:

* Has its own conversation history
* Can be assigned [tasks](/tasks)
* Can attach to a notebook to write and run research code

Agents appear in the **Agents sidebar** (chat bubble icon in the activity bar).

## Starting an agent

Click the **New Agent** button in the top left to create a new agent. You can also create agents from within a conversation when you need parallel work.

Execution mode applies to every agent. Choose **Notebook** in **Settings > General** for interactive notebook work or **Script** for standalone files and shell-driven jobs. Agents created by other agents inherit the same mode.

## Parallel agents

The biggest time sink in research is sequential experimentation — you try one thing, wait, check results, then try the next. Qualia eliminates that by running multiple agents in parallel.

### Example: model comparison

Ask Qualia:

> *"Compare three approaches: (1) random forest with default hyperparameters, (2) XGBoost with early stopping, and (3) a small neural network. Use 5-fold cross-validation for each. Run them in parallel."*

Qualia creates three agents, each working on one approach simultaneously. When they finish, you can compare results side by side.

Use the **Parallelism** research-style control in chat or under [Settings > AI](/settings/ai) to choose how readily Qualia splits suitable work across agents. Low usually keeps work in one agent, Medium delegates when useful, and High proactively looks for parallel work.

### Example: hyperparameter sweep

> *"Try learning rates of 0.001, 0.01, and 0.1 with the current model. Train each for 50 epochs and compare validation loss curves."*

### Example: feature engineering

> *"Test three feature sets: (1) just the raw numeric columns, (2) with polynomial features, (3) with target encoding on categoricals. See which gives the best F1 score."*

## The Agents sidebar

The **Agents sidebar** shows all your agents:

* **Status indicators**: Running (spinner), completed (checkmark), paused, stopped, or waiting
* **Hierarchy view**: See parent-child relationships between agents
* **Time grouping**: Agents grouped by when they were created
* **Pending confirmation**: Agents waiting for approval (when confirmation is required)

### Agent actions

From the sidebar, you can:

* **Open**: Jump to an agent's conversation
* **Pause/Resume**: Temporarily stop or continue work
* **Stop**: Permanently stop the agent
* **Delete**: Remove the agent and its history

## Agent communication

Agents can communicate with each other:

* **Message agent**: One agent can send a message to another
* **Agent links**: Click to navigate between related agents in chat
* **Shared knowledge**: Agents in the same project share the [Knowledge System](/knowledge)

## Agent confirmation

When an agent wants to create a new parallel agent, it can require your approval first. This is controlled in [Settings > AI](/settings/ai):

* **Enabled**: New agents wait for you to approve
* **Disabled**: Agents start immediately

When enabled, pending agents show in the Agents sidebar and you can approve or deny them.

## Integration with tasks

Agents work on [tasks](/tasks):

* Tasks can be assigned to specific agents
* Agents work through their task queue
* You can reassign tasks between agents in the [Gantt view](/tasks)

See [Tasks](/tasks) for details on the task system.

## Autonomous work

For extended independent work, see [Autonomous Mode](/autonomous-mode). With **Autonomous Mode** on, agents work without interruption and report progress via Slack.