One rulebook for your team and your AI agents.

For agencies and companies running Claude Code seats and AI agents across several clients or teams. Practical Works sets up one shared brain with your team and keeps it maintained. Everyone loads the same rules at the start of every session, each client gets its own, and agents can propose changes but never slip them in.

The problem

  • Rules that disagree

    Every seat has its own CLAUDE.md, and by the end of the month they disagree.

  • A rule the agent never saw

    An agent shipped something that broke a client's rule it had no way of knowing.

  • Lessons that stay put

    A good lesson from one project never reaches the next one.

How it works

  1. We assess how you work today

    On a call we map your seats, agents, clients and the rules each one follows now, and where those rules already disagree.

  2. We set up the brain with your team

    Your doctrine, skills and client rules become notes in a shared vault you own. We connect every Claude Code seat and every agent so each one loads the current version when it starts. Nobody pastes context.

  3. We keep it maintained every month

    Agents propose what they learned. A person on your team approves it, the ledger records who changed what, and any change can be reverted. We run that upkeep on a monthly retainer.

Proof

  • Practical Works runs its own shop this way: people and agents work under one constitution served from the vault, and more than 70 skills load from it into Claude Code seats and our PM agent.
  • Every change an agent makes to that constitution is a ledgered promotion. It names who changed what and when, and any one of them can be reverted.
  • Three separate brains, each for a different business, run on the same library. One of them runs daily operations at Breezy Keys, our property business in Tampa Bay.

Frequently asked questions

Is this software we set up ourselves?

No. It is a Practical Works engagement. We start with an assessment call, then set up the brain with your team, connect your seats and agents to it, and keep it maintained on a monthly retainer. We quote the price once we know your clients and tools.

Who is this for?

Agencies and companies with several clients or teams that already use AI tools such as Claude Code, Cursor or their own agents, and want all of them working from the same rules.

Why not a shared CLAUDE.md in git?

That works for one repo and one client. It breaks when you have several clients, agents running outside your laptops, and a need to see who changed a rule.

Where does the data live?

In a Google Shared Drive you control, as plain markdown files. You can read every note without us.

Which agents work with it?

Claude Code seats, and agents that can call a library or an MCP tool. Ours include a Mastra agent and an open-source Hermes PM agent.