A brain and hive mind

Plexus Hive Mind

design stage · personal + team infra

What: One brain per user. On a schedule it reads that person's email, messages and Claude Code sessions, and keeps both private and public persistent records. Privately, a daily summary and indexed notes into an Obsidian markdown vault. Publicly, documentation for teammates and leadership - a transitory log carrying the day's blockers and contributions, and permanent doc updates for every feature change. Across a whole team it acts as a hive-mind tracking project progress and blockers.
Why it was hard: These insights tread a fine line of Big Brother level surveilance that encroaches on a user's autonomy. Finding that balance of useful information while not adding anxiety to a worker's day to day is the most important part.
What I decided: Splitting int public and private files gives control back to the worker, but still allows authorized oversight of projects. This model essentially keeps docs up to date autonomously.
What happened: This is implemented across dozens of teams at Capital One with over 250 daily users as well as in my own personal work.

All the data is ready for scraping, but it lives scattered across inboxes, git branches, and communications with teammates so the only way to understand exact project progress is attending team standup.

Plexus is the attempt to make real time info indexable without turning everyone’s private correspondence into company documentation.

The shape

four sources in, three records out — the vault is yours alone, the dotted lines are the only things the brain shares, and the return path is the hive: the docs you read are the ones every other brain is writing

One brain per person, running on their own machine. Each brain has access to public docs which are being daily updated by teams autonomously. Docs are further divided using role based access control.

The brain

Plexus aggregates information that is already public into one place. It does the work nobody remembers to finish and keeps everything in order. It updates docs and summarizes project status. This way LLMs can easily get status without having to build a map to all related project information and spending massive tokens.

As a private brain, Plexus helps each worker perform their normal day to day work and provides a persistence layer to learn more about each workflow and make it more efficient. It helps a user prep for meetings, create presentations, and adapt designs as more information is uncovered.

The hive mind

The second wave a benefits comes from a leadership and synergy perspective.

Each design meeting creates meaningful updates on a project status, not just in the minds of the attendees. Any other permissioned user has immediate access to decisions without having to ask for a summary.

If a development happens while a dev is on vacation, they are immediately notified of all changes and why when they come back.

How is this different from CLAUDE.md

Claude and OpenAI agents have built in persistence.

This memory is only managed through their app use and not through bedrock LLM instances (which most devs use for cost and licensing reasons)
Instead of injecting into every conversation as context noise, the brain registers as a skill that the session can call on to inject specific context based on indexed results. The brain can grow into the GB and context remains relevant.