Celeste for law firms: intake that screens, status that answers

Two kinds of messages fill a law firm’s inbox, and they waste attorney time in opposite ways. New inquiries arrive raw: a long story, no dates, no sense of whether it is a case the firm takes. Existing clients write in with a question that deserves an answer and needs none of an attorney’s judgment: has the other side responded, when is the hearing, did you get my documents.

Celeste handles both, on email and web chat, without practicing law at anyone.

Intake that arrives pre-worked

A prospective client who emails the firm or opens the chat box gets engaged immediately, at 9pm on a Sunday if that is when they write. The agent asks the questions your intake sheet asks: what happened, when, where, who else is involved, what deadlines might be running. What the attorney receives is not a forwarded wall of text but a structured summary with the facts your screening actually turns on.

The screening criteria themselves live in a playbook, which means the agent applies them without disclosing them. If the firm declines matters below a certain value, or conflicts out of a certain industry, the prospect never learns the threshold. They get a courteous outcome; the reasoning stays yours. This is enforced, not requested: outgoing drafts are scanned for playbook content before send, and even a staff reply that quotes a private procedure is stopped until someone deliberately confirms it.

The agent gives no legal advice. That is not a tone setting, it is a topic boundary. Substantive legal questions are outside every topic you allow, and a message that wanders into one is held for a human. Words you never want answered autonomously, “malpractice,” “bar complaint,” “subpoena,” go on an escalation list, and a message containing one is barred from an automated reply outright.

Status without the interruption

For current clients, connect your practice management system as a read-only tool. A client who writes in verifies their identity first, and every lookup the agent runs is scoped to that client’s own matters by code outside the AI. Client A cannot ask about client B’s case, and neither can someone spoofing client A’s email, because unverified senders never get an autonomous reply containing record data.

Within that scope, the routine questions answer themselves. Did the filing go in. When is the deposition. What documents are still outstanding. Each one is a two-minute interruption the paralegal no longer absorbs, and the client gets the answer in minutes instead of whenever the office surfaces.

Trust on the firm’s schedule

Nothing answers autonomously on day one. The agent starts in shadow mode, drafting silently against your real inbox while you read what it would have said. Firms typically graduate logistics first, office hours, document delivery, appointment mechanics, then status lookups once the identity scoping has proven itself, and leave intake summaries permanently in a mode where a human sends the final word. Every topic has its own dial, and any of them can stay at “draft only” forever.

A law firm’s product is attorney judgment. Every hour spent retyping a hearing date is that product going unsold. The queue Celeste takes is precisely the part that never needed the judgment, and what your team receives is triaged, summarized, and scoped.

Read how it works for the practice-first model, or see the product. The trial runs in shadow mode against your live inbox, so you can judge the drafts on your own matters before anything sends.