Skip to main content
Open Settings > Agents to set defaults for new agents, decide what they can run without asking, and choose how they notify you. An agent is a chat that works on your behalf; see Agents for how agents run and delegate work.

Research style

Research style is a set of controls that shape how an agent approaches your request. The values under Defaults for new agents apply to every agent you start from then on. Every control starts at Medium. To change one chat without touching your defaults, click Research style in the chat box. A change you make while the agent works applies from its next step. Agents that another agent creates inherit some of these values; see Parallel agents.

Curiosity

Curiosity sets how far the agent explores beyond your explicit request.

Independence

Independence sets how often the agent uses its own judgment instead of asking you. When you choose Auto ∞, Qualia dims Curiosity and Parallelism because autonomous mode ignores them. Auto-approve commands shows as on, because Auto ∞ already includes it.

Parallelism

Parallelism sets how readily the agent splits work across additional agents.

Verbosity

Verbosity sets how much detail the agent writes in replies, findings, and reports. It doesn’t change how much research the agent does.

Auto-approve commands

Turn on Auto-approve commands in a chat’s Research style menu to let the agent run shell commands and scripts without an approval card. The agent keeps your Independence level, so it can still ask you questions. Library installs, MCP tools set to Ask each time, and recursive deletes of a drive root or your home folder still ask first. Auto-approve is off by default. To turn it on for every new agent, choose Run everything under When agents run commands.

Other defaults for new agents

Command approvals

When an agent wants to run an untrusted shell command or script, a card above the message box shows you the exact command.
  • Click Run to run it once.
  • Click Skip to decline. Qualia tells the agent, and the agent continues.
  • Type a message instead to cancel the command and send your message as new instructions.
  • Open the arrow beside Run to always allow the command. If it has several parts, check the prefixes to trust, or check All, then click Run.
Library installs and new-agent requests use the same Run and Skip card.

When agents run commands

Under Settings > Agents > Permissions, choose the default for new agents: Chats you’ve already started keep their own auto-approve setting. While your default Independence is Auto ∞, Qualia locks this choice to Run everything.

Always-allowed commands

Expand Always-allowed commands to see the command prefixes you trusted with Always allow & run. These run without asking, whichever option you picked above.
  • Add a prefix yourself, such as git status, or remove one to require approval again.
  • Approvals are specific to your operating system.
  • Risky flags still ask first, such as -rf on rm.
  • Shells and inline scripts can’t be always allowed, because that would let any command run unattended. For example, approving python script.py doesn’t approve python -c "...".

Other permissions

  • Auto-approve GPU requests (Qualia Cloud): lets agents add GPUs without asking, up to an hourly price. Choose Always ask, $2/hr**, **$10/hr, $50/hr, or Any price. The default is Always ask.
  • Sandbox limits: set the folders agents can edit, the files they can never edit, and the sites their commands can reach.
  • Computer use (desktop, experimental): lets agents open and use apps on your computer. You approve each app the first time a chat uses it, unless the chat is autonomous.

Search and chat

Web search with your own keys

If you use your own OpenAI or Anthropic key under Settings > Models > Provider access, choose how agents on those models search under Web search with your own keys:
  • Provider search (default): agents search through the provider, under your agreement with it.
  • Qualia search: every model searches through Qualia.
The control appears only on bring-your-own-key accounts; see API keys.

Notifications

In a browser, including Qualia Cloud, click Enable notifications under Browser notifications, then choose Allow in your browser’s prompt. Qualia also offers this when an agent first asks for your input. After you allow it, click Send test notification to check delivery.
Browser alerts arrive only while the Qualia tab stays open and connected in a supported desktop browser. Grant permission separately in each browser you use. If alerts don’t appear, check your browser’s site settings and your system’s Focus or Do Not Disturb settings.

Advanced

Agent models

Set agent models under Settings > Models:
  • Agent Model is the model for the agents you start. If you never choose one, Qualia uses Auto, which picks a strong model for you.
  • Sub-agent model, under Advanced model overrides, is the model for agents that other agents create. By default, these agents use the creating agent’s model (Use parent agent). To pick a separate model, turn off Use parent agent. A faster, cheaper model often suits these focused tasks.
Changing Agent Model also changes existing chats, unless you picked a model in that chat’s model selector. A chat with an explicit pick keeps it until you change it there. To pick a model for one chat, see Choose a model.