Anthropic Tests 'Conway' — An Always-On AI Agent That Lives on Your Computer

Anthropic is testing a new kind of AI agent internally. It's called Conway, and it differs fundamentally from anything the company has launched before.

Where Claude today is a tool you open and close, Conway is designed to always be present. It runs in the background, observes its environment, and acts on its own initiative — without waiting for you to ask.

The details emerged through an accidental leak of Claude Code source code on GitHub in late March. Anthropic confirmed it was human error, not a security breach.

According to the leaked material, Conway features an independent sidebar UI, deep Chrome integration, and webhook support. That means the agent can be woken by external events — an incoming email, a system update — and act autonomously. Users can install custom extensions via a new package format (.cnw.zip).

The source code also revealed references to a system called KAIROS — a three-layer memory architecture that lets Conway consolidate information and maintain context over time, even while the user is idle. The idea is to build a "digital twin" that understands what you're working on without needing you to explain it again.

The ambition is clear: Anthropic doesn't just want to build a better chatbot. They want to build a new operating system for AI.

Conway is not yet launched and no release date has been announced. But it's evident that the company is well advanced in development — and ready to flip the switch when the time comes.

For CIOs and technology leaders, this is an important signal. The time horizon for always-on AI agents in enterprise environments is no longer years — it's months.