Beacon gives Claude Code and 20+ agents shared memory

Coding agents have so far worked in silos: corrections made in Cursor do not carry over to Claude Code, and skills built in Codex do not transfer to other harnesses, so switching tools means starting from scratch. Beacon addresses that by using OpenTelemetry, hooks and local session files to normalize every prompt, tool call, command, file change and response into a single structured trace, regardless of which agent produced it.

The demo showed two practical consequences. First, continuity: if a user hits a usage limit in Cursor, they can open Claude Code and ask it to continue the latest Cursor session while Beacon retrieves the relevant context in milliseconds. Second, skill creation: Beacon identified a recurring UI development workflow and generated a skill containing the exact dev server commands, screenshot steps and deployment instructions. It also analyzes usage across thousands of sessions to flag skills worth keeping, combining or removing.

The security side is where the demo turned concrete. Beacon ships with hundreds of detection rules that work across agents, including sequences where an agent reads sensitive credentials and then makes an outbound network request. Reviewing Justin's actual Cursor history, it surfaced an incident in which part of a production database password had been sent to a model provider — an example of the visibility the layer is meant to provide.