Abilities

Harness is centered on one core loop: understand the screen in front of you, then recover the right context later without making you restate it. Everything else builds on that.

Start here

• Screen awareness: current-screen grounding, tab audio, and private historical recall.
• Memory: the on-device memory database and how relevant context enters a turn.
• Chat & the PiP: the floating window, steering, and background tasks.
• Models: the on-device models, and who runs your AI: managed, your own key, or your own endpoint.

Getting things done

• Watch & catch: watch a screen for you and flag the moment something happens.
• Automations: work on a schedule toward something you asked for.
• Files, images & video: read your files and make documents, images and video.
• Skills: the skills library and skills of your own.
• The deep-agent brain: planning, research and verification for hard questions.
• Connections (MCP): let other AI tools ask your Harness memory, and give Harness tools from MCP servers.

Beyond the screen, the extension brings Harness into the page you are on, and Harness Relay hands work to the agents you already use.