Praxis has a connector. Point Cursor, Claude, ChatGPT, Grok, or Gemini at it, approve the connection once, and the assistant works with your case files directly. Not a chat window bolted onto the product — the same platform, the same rules, from wherever you already type.
The interesting part is not that an assistant can read your data. It is what it cannot do.
It sees what you see
The connector does not get an all-access key. It acts as you, in one organization. An agent who cannot open Legal’s cases in the interface cannot open them through an assistant either. You disconnect it when you want. Two organizations are two connections.
What you can ask
- What is in my inbox, and what has been sitting there longest?
- What is missing to close PRAXIS-9QK4TZ?
- Explain this procedure
- What happened on these five cases since Friday?
- Who still has to sign?
Those answers are not guesses. The assistant does not invent a step that does not exist in your flow, and it uses the names your organization gave things.
It can also do the work
With the matching permission, the assistant opens a case, applies a step, leaves a note, attaches a document, or picks up a task from your inbox.
The one that surprises people is onboarding. Hand it the staff spreadsheet; it asks when a role or an area is unclear, and it sends the invitations. It never sets a password — people still choose their own when they accept.
The trail survives it
Everything done through an assistant is marked as such. Months later you can still tell what a person did in the interface from what an agent did on their behalf. An assistant cannot rewrite yesterday, because nothing can.
If this looks like a problem you have, write to us.