Stop re-explaining your clients to AI

Krewbill keeps a shared, approved memory for each client, so your whole team gets the same tone and rules in Claude, ChatGPT and Cursor.

The problem

Every new chat starts from zero.

  • A new chat means pasting the tone rules again

    The assistant does not remember how a client likes things written, so someone pastes the same background into every conversation.

  • Each teammate pastes a different version

    One person has the old brand guide and another has last month's rules. The work stops sounding like one agency.

  • Freelancers start from zero

    Someone new to a client has none of its history, so they ask around or guess.

How it works

Three steps.

  1. 01

    Create a space for each client

    Every client gets its own space, with its own people and roles.

  2. 02

    Add notes, and approve them

    Write down the tone and rules, key facts and decisions. A person approves each note before an assistant can read it.

  3. 03

    Connect Claude, ChatGPT or Cursor

    Make a key for the spaces you choose and add it to your tool over MCP. The assistant reads the approved notes before it answers.

See the connection guide

Before and after

What changes in a day of work.

Before

  • Tone rules pasted into each new chat
  • Teammates working from different versions
  • A new freelancer starts from zero
  • A client's rules live in someone's head

With Krewbill

  • One approved memory for each client
  • Every connected assistant reads the same notes
  • A freelancer you add to a space sees the same notes
  • A change is suggested, and a person approves it

Safety

Assistants suggest. People decide.

  • Assistants can only suggest

    A suggestion waits as pending. An assistant cannot add a note to the shared memory by itself.

  • A person approves

    Only people with approval rights can approve or reject a note. Assistants never can.

  • Each key is limited to the spaces you choose

    That can be a single client. A key can also be read only, and you can revoke it at any time.

How security works

Available today

What you can use now.

  • Accounts, workspaces, spaces, roles and invitations
  • A space for each client, with notes grouped as voice and rules, key facts and decisions
  • Connection for any AI tool that supports remote connectors, with keys limited to chosen spaces
  • An approval screen where assistants' suggestions wait for a person
  • Keyword search, export to Markdown or JSON, and deletion

Krewbill is free during early access.

Keep passwords and other secrets out of your spaces, and start with a low-risk client.

Questions

Plain answers.

Does it work with ChatGPT?
+
Krewbill connects to ChatGPT, Claude and Cursor as a remote MCP server. It works when your tool and plan let you add a custom connector. What you can add depends on your tool, your plan and your settings, and we have tested it with only a few tools. The connection guide has the steps for each.
Who sees the notes?
+
Only the people you add to a space, each with a role: viewer, editor or approver. Owners and admins of the workspace can see every space. An assistant sees the approved notes of the spaces its key was given, and nothing else. Anyone who holds a key can read those notes, so keep keys private and revoke any that leak.
Can the AI change notes by itself?
+
No. An assistant can only suggest a note, and the suggestion waits until a person approves or rejects it. It can also flag a note as possibly out of date, which a person reviews. It never edits, approves or removes a note.
What does it cost?
+
Krewbill is free during early access. If that changes, we will tell you before anything is charged.

Try it with one client.

Make a space, add a few notes, and connect your AI tool.