Skip to main content
Let an agent work through a long research task without stopping for your approval, and follow its progress from Slack or the Chats sidebar. Autonomous mode is the Auto ∞ level of Independence, one of the research style controls.

What changes in autonomous mode

In autonomous mode, the agent makes its own decisions instead of waiting for you:
  • It runs commands without approval cards, as if Auto-approve commands were on.
  • It doesn’t ask you questions. It proceeds on its own best judgment.
  • It ignores Curiosity and Parallelism, so Qualia dims those controls.
  • Agents it creates also run in autonomous mode.
Qualia still asks before a few destructive commands, such as a recursive delete of a drive root or your home folder.
Don’t use autonomous mode in a workspace holding data you can’t afford to lose, because the agent runs commands without showing them to you first.

Turn on autonomous mode

  1. Click Research style in the chat box.
  2. Under Independence, choose Auto ∞.
To make it the default for new agents, choose Auto ∞ under Independence in Settings > Agents. To leave autonomous mode, choose Low, Medium, or High. If you fork an autonomous chat, the fork doesn’t keep Auto ∞, so its commands ask first. See Fork a chat.

Get updates in Slack

  1. Connect Slack under Settings > Integrations > Slack.
  2. Open the chat’s … (Chat actions) menu, or right-click the agent in the Chats sidebar, and choose Notify in Slack.
The agent posts progress updates, preliminary findings, and completion notices to that chat’s Slack thread. Reply in the thread to steer it. For example, propose a new direction, ask it to focus on one approach, or request more detail on a promising result.

Start an autonomous run

  1. Choose Auto ∞ as described above.
  2. Describe a research direction, for example:
    “Explore different architectures for this time series forecasting problem. Try at least 5 approaches and report the best.”
  3. Leave the agent to work. It posts to Slack if you turned on Notify in Slack.

Monitor a run

While an autonomous agent runs, follow it from three places:
  • Slack: progress updates, findings, and completion notices in the chat’s thread.
  • Chats sidebar: the agent hierarchy, activity indicators, and the ideas agents are testing, with their findings and metrics. See The Chats sidebar and Ideas and tasks.
  • Knowledge: the claims and evidence agents recorded. Click Knowledge (lightbulb) in the titlebar, or choose View > Knowledge. See Knowledge.

When to use autonomous mode

Autonomous mode suits work you’d otherwise have to supervise step by step:
  • Open-ended exploration, such as finding the causes of every anomaly in a dataset.
  • Model sweeps, where the agent tries many configurations and follows up on promising ones.
  • Long-running experiments, such as training runs to monitor for hours and stop when they stop paying off.
Scheduled pipeline runs always run autonomously, because nobody may be around to answer.