Work in parallel. Stay in the loop.

Run coding agents, review their decisions, and pick up the work from one shared browser workspace.

Decide before the next change.

An agent can inspect the project and propose a plan before editing. Send it back with feedback, or approve the approach.

• Choose automatic execution, confirmation for write and execute tools, or review of every tool call.
• Keep prompts, decisions, plans, and tool results in session history.
• Use hooks for project checks and skills for repeatable instructions.

Your machine. Your model connections.

Use a personal daemon or shared company machine, with local workspaces, Git worktrees, or Docker sessions. Closing the browser does not stop an agent. The execution machine must stay awake and running. Model requests and session history can contain code snippets and tool output.

Connect provider API keys, supported subscriptions, Azure deployments, or local OpenAI-compatible endpoints. Provider costs stay separate. AFK adds no token markup.

Make good workflows repeatable.

• Delegate focused work to sub-agents.
• Connect external tools through MCP.
• Use scheduled automations and inbound webhooks, subject to your plan.
• Search earlier sessions and resume with context.