Hey Clicky and the interface that points

Most AI assistants still make people leave the work, open a blank chat, explain the context, and then translate the answer back into action. Hey Clicky is interesting because it flips that interaction. The companion lives beside the work, can speak, observe screen context, and point to the thing it is talking about.

See Hey Clicky, watch the original Clicky demo, and explore the public implementation.

What is genuinely good about the idea:

• Context is visible. A person does not have to describe every panel, field, and workflow before asking for help.
• Guidance is spatial. Pointing at the next action is often clearer than a paragraph of instructions.
• The assistant has a presence without becoming the entire interface. It can be quiet until help is needed.
• The interaction has states. Listening, thinking, speaking, and pointing tell the user what is happening.

The public Clicky repository is also a useful architecture lesson. It separates the on-screen companion, transcription, model response, voice output, and a secure server-side proxy for keys. The visual charm is only half of the product. The other half is clear boundaries: what the assistant can see, what it can do, and when it must ask before acting.

For enterprise software, this pattern has real potential. Imagine an assistant that can highlight the exact variance driver, show the source transaction path, explain a control exception, or walk a new analyst through a planning workflow. The best version does not cover the work with a giant modal. It makes the next useful thing easier to see.