Skip to main content
Open Settings > Agents to configure agent behavior and defaults for new chats.

​
Browser notifications

In Qualia Cloud, open Settings > Agents > Behavior and click Enable notifications under Browser notifications. Choose Allow in your browser’s permission prompt. Qualia also offers this setup when an agent first asks for your input.
  • Notify when an agent needs input alerts you when an agent asks a question or needs command approval, including approval to run Python, while Qualia is unfocused. Click the notification to return to that agent’s chat.
  • Notify when agents finish alerts you when all agents in the workspace have completed while Qualia is unfocused.
  • Agent interaction sounds controls the accompanying sounds separately.
Once permission is granted, Send test notification lets you check delivery. If notifications are blocked, allow them in your browser’s site settings and return to Qualia. Also check your operating system’s notification and Focus / Do Not Disturb settings if alerts are not appearing.
Keep the Qualia tab open and connected in a supported desktop browser. Notifications can appear while you use another tab or application, but are not delivered after you close the tab or while it is disconnected. Browser permission must be granted separately in each browser you use.
Enable or disable web search. It works with every provider:
  • Enabled: AI can look up documentation and references
  • Disabled: AI works only with local context
Web search is useful for staying current with APIs and best practices.

​
Web search with your own keys

By default, searches run through Qualia search. On a bring-your-own-key account, if you route OpenAI or Anthropic directly through your own API key under Settings > Models > Provider access, agents on those models search through the provider’s own web search instead, using the same model as the chat, so searches stay under your agreement with that provider. Routes with a custom base URL always use Qualia search. If the provider refuses the search (for example, your organization has web search turned off), the search falls back to Qualia search and the conversation continues. A Web search with your own keys control appears at the bottom of Provider access when this applies:
  • Provider search (default): OpenAI and Anthropic models on your own key (without a custom base URL) search through the provider’s own web search
  • Qualia search: searches route through Qualia for every model
Accounts without bring-your-own-key, and models on the Quadrillion proxy, always use Qualia search and do not show this control.

​
New agents

Settings for how agents are created and configured.

​
Default model for agent-created agents

When an agent creates parallel agents, this model is used:
  • Consider using a faster, cheaper model for parallel work
  • Parallel agents often handle focused, bounded tasks

​
Default model for user-created agents

When you start a new chat or agent, this model is used:
  • Your primary model for direct interaction
  • Usually your most capable model
  • If you never choose one, Qualia starts you on Auto, which picks a Claude model for you, and falls back to another provider only when Claude is unavailable
Changing this default also applies to your existing chats, unless you have explicitly picked a model for a chat from its model selector. A chat with an explicit pick keeps that model until you change it there.

​
New agents require user confirmation

Control whether agent creation needs approval:
  • Enabled: You approve each new agent before it starts
  • Disabled: Agents start immediately
Pending agents show in the Agents sidebar when confirmation is required.

​
Research style

Default curiosity, independence, and parallelism for new agents.

​
Curiosity

How much the agent explores beyond your immediate request:

​
Independence

How much the agent assumes versus asks:

​
Parallelism

How much the agent splits work across additional agents:

​
Auto-approve commands

When on, new agents run shell commands and scripts without asking for approval but can still ask you questions. Independence Auto ∞ already includes this, so the switch is locked on while Auto is selected.

​
Autonomous Mode

Choose Independence → Auto ∞ to run without confirmations. Connect Slack and enable Notify in Slack in the chat actions menu if you want progress updates there. See Autonomous Mode.

​
Per-session overrides

These settings are defaults. You can adjust curiosity, independence, parallelism, and auto-approve commands for individual chat sessions without changing the global settings.

​
Rules

Persistent instructions that guide AI behavior. See Rules for full documentation. In Customize agent, you can:
  • Create rules: Add new rule files
  • Edit rules: Modify existing rules
  • Import rules: From Cursor or Claude Code
  • Reveal folder: Open rules directory in file manager

​
Skills

Reusable workflows the AI can invoke. See Skills for full documentation. In Customize agent, you can:
  • Create skills: Add new skill files
  • Edit skills: Modify existing skills
  • Toggle auto-invoke: Control whether agents can trigger skills automatically
  • Import skills: From Cursor or Claude Code

​
Bash command allowlist

When Qualia wants to run a shell command or script that isn’t trusted yet, a confirmation card appears above the message box showing the exact command. Choose Run to run it once, open the Always allow & run dropdown to trust a command prefix, or Skip to decline. Skipping tells Qualia you declined and lets it continue; typing a message instead cancels the command and sends your message as instructions. Library installs and new-agent approvals show the same Run / Skip card. Commands you’ve approved with “Always allow & run” appear here:
  • Approved commands run without confirmation in future
  • Remove commands to require confirmation again
  • Approvals are specific to your operating system.
  • Shells and command wrappers still ask for confirmation because they can run arbitrary commands.
  • Approving an interpreter command such as python script.py does not approve inline scripts such as python -c "...".