Compliance with Celeste: Trust, not just technology

Moving faster can’t mean losing control of the decisions that protect the firm.

That tension came through clearly in the questions we received after our Compliance with Celeste: First Look webinar. Firms wanted to know what Intapp Celeste can do, but just as importantly, how they stay in control while it does the work. They asked which models sit behind it, what it can access, and where a person still makes the call.

Those are fair questions to put to any vendor, so I’ll work through a few of them here.

What Celeste really is

Much of the attention on AI in legal has gone to two kinds of tools: general-purpose models that lawyers use for everything, and tools built to help with the practice of law. Celeste is neither. It’s Firm AI, built for the business of law: how a firm opens matters, clears conflicts, tracks obligations, and runs the operational side that never shows up in a brief but certainly shapes how you serve your clients.

That distinction changes the stakes. A drafting suggestion gets read by the lawyer who asked for it. A system working with client data, conflicts records, and outside counsel guidelines on your firm’s behalf needs controls built around it.

I think about Celeste in three layers. Chat is the simplest: a better way to ask a question and get an answer from your own data instead of digging through systems. Assistants go further: you ask for something specific, and Celeste does it. Coworkers go furthest: you give an agent a defined job and instructions, and it does that work. Whether you’re asking a question, delegating a task, or assigning an agent a defined job, each layer depends on trust in your data, your processes, and the controls around them.

Who picks the model

Some attendees asked which AI models sit behind Celeste. Your firm’s administrator picks the preferred model for the job and can switch anytime. I think that’s the right call. Models are evolving so fast that tying a firm to a single option is a bad bet, and a compliance function shouldn’t depend on one model’s strengths, weaknesses, and roadmap. A firm that can choose the right model for the task will be more successful.

How Celeste works within existing permissions

As we explained in the webinar, Celeste uses the requesting user’s credentials to perform tasks on their behalf. It’s designed to work within the firm’s existing permissions and access controls, not around them. In the very near future, Celeste will have coworkers that can have their own credentials, so they can truly operate autonomously within their own permissions and controls.

What that looks like in practice

The three examples we showed in the webinar put those ideas to work.

In the first, a conflicts analyst needed to evaluate a potential engagement involving multiple parties. A playbook, a set of plain-language instructions built around the firm’s process, handled the setup. It built the search strategy, identified related entities, ran the search, and organized the results. The analyst spent their time evaluating the important, sensitive hits that required business decisions, not managing the mechanics of the search.

In the second, a partner received an email with instructions for a new matter. Celeste used it to prepare the request in Intapp Intake, and a person reviewed that request before anything was submitted. The work shifted from retyping the request to checking it.

In the third, a partner preparing for a client call needed to know whether an advance waiver was in place. Celeste pulled the exact clause from Intapp Terms and linked back to the record, so the partner had the agreed language in front of them instead of a summary of it. It also identified what still needed clarification, so the partner could see where their judgment was required. That’s the difference between an answer you take on faith and one you can check against the source.

In all three, Celeste did the assembly and a person kept the decision.

For compliance teams managing growing workloads, that means less time gathering and organizing information, and more time evaluating risk and making decisions.

Who’s accountable for the agent?

Handing work to Celeste doesn’t hand off responsibility for it. When Celeste acts on behalf of a user, its actions are recorded in the activity logs because it performs the actions within the applications. Celeste is working within the bounds of the guardrails you’ve defined in Intapp Walls policies, playbooks, connectors, and application permissions. The analyst still owns the conflicts decision, the partner still owns the matter, and the firm still owns the outcome.

Your process, in plain language

The conflicts example ran from a playbook built around the process a firm wants followed. That matters in compliance, where client commitments, regulatory obligations, and internal policies differ from firm to firm, and the exceptions count as much as the standard process. AI has to fit how a firm manages those obligations. It can’t ask the firm to adopt a new way of managing them.

More capacity, without compromising control

The opportunity with Firm AI is to give compliance teams more capacity without asking firms to trade away the controls and professional judgment that make those teams effective.

AI changes how compliance work gets done. It doesn’t change who answers for it. That’s the bar we’re building Celeste against, and it’s the bar I’d hold any vendor to if I were sitting where you are.

See it in action. Watch the on-demand recording of our Compliance with Celeste: First Look webinar to see all three examples. Then visit the Compliance with Celeste product page to learn more. Have questions about how Celeste would work in your environment? Connect with your Intapp account team.