Purpose
Referent is an AI-native legal practice management software, aimed at law firms. Its positioning is not as a traditional "system of record," but as a "system of action": legal AI agents carry out daily operational work, while lawyers approve key actions, thereby devoting their time to clients, relationships, and business growth.
It is officially stated that its team is supported by Google for Startups, the Nvidia Inception Program, and the OpenAI Select Partner program.
Main Capabilities
According to official information, Referent's agents can cover the following operational areas:
- Run the business through chat: there is no need to manually enter items one by one as with traditional legal software; simply tell Referent what needs to be done, the agent executes it, and the lawyer approves it.
- Client intent and intake: clients converse with AI through an intake link, Referent assesses fit, opens a case file, and prepares engagement materials for lawyer approval ("Send a link. Sign a client.").
- Email filing: after connecting an inbox, Referent files emails, attachments, and context into the corresponding case files, and flags content that requires lawyer attention.
- Document drafting: based on the full context of the case file, it pre-generates drafts of replies, letters, and documents, which the lawyer edits, approves, and sends.
- Deadline tracking: monitors court dates, filing deadlines, statutes of limitations, etc., and reminds before they become urgent.
- Billing: records time during work, automatically compiles invoices (hourly or flat fee), and lawyers review bills rather than timesheets.
- Follow-up: every outreach, such as payment reminders, client check-ins, and review requests, is pre-drafted and will not be sent without approval.
- Approval of key steps: any send, filing, or billing action requires the lawyer's signed confirmation, and all operations are logged and auditable.
- Workspace migration: migrates clients, case files, tasks, and documents from existing tools into Referent, enabling immediate use from the start.
- Mobile: provides iPhone and Android apps, allowing users to view agent work at any time, approve key actions, and review case files between court sessions.
- MCP server: follows the open standard Model Context Protocol and can connect to ChatGPT, Claude, Perplexity, Hermes, Gemini, OpenClaw, and other MCP-compatible applications handle clients, case files, and documents in chat, with permissions, approvals, and audit trails kept consistent.
How to Use
The official description's core mechanism is "AI prepares, lawyer approves": the agent does the legwork and places results in a pending-review state rather than disposing of them on its own. Any client-facing or high-risk action requires the lawyer's explicit sign-off, every action is recorded in an inspectable audit trail, and every draft can be edited or rejected before it is sent.
According to the official FAQ, becoming AI-native happens in three steps: first, connect the firm's context by bringing clients, case files, emails, documents, and deadlines into one system; second, hand repetitive operations to AI agents, typically starting with the seven workflows of intake, follow-up, task creation, deadline tracking, document filing, billing preparation, and client updates; third, keep approval authority in the lawyer's hands, setting rules to ensure that client-facing or high-risk actions do not occur without review. Officially, Referent provides white-glove onboarding to complete setup and connection work on the firm's behalf, enabling the firm to reach a usable AI-native baseline within days.
Referent will request to connect Google services, and only connects the services the user selects: Gmail for filing client correspondence into the correct case file and sending approved replies; Google Calendar for two-way syncing of hearings, meetings, and deadlines; Google Drive for making connected documents searchable within the relevant case files. Officially, these access permissions are granted to the extent required by each feature, Google data is used only to provide these features, not to train AI models, and not for advertising, and any service can be disconnected at any time, after which synced data will be deleted.
Use Cases
Officially, three types of users are listed:
- Solo Lawyers: operate like a modern law firm as a team of one, with AI agents handling routine work.
- Boutique Firms: equip every lawyer with standardized workflows and agents that automate operations.
- Scaling Operations: deploy an AI operations layer across multiple practice groups, with every action logged, auditable, and approved by a lawyer.
Data and Security
Officially stated: never use firm data to train AI models; never sell or monetize client information; encrypt and isolate workspaces by firm, without sharing data between clients; deletion means completeDelete. Control and oversight: key operations require lawyer approval, and every AI operation is logged and auditable. Access security: role-based access for authorized firm members; AES-256 encryption at rest, TLS 1.2+ in transit; SOC 2 and ISO 27001 are in preparation.
Differences from general-purpose chatbots
According to the official description, a blank AI chat knows nothing about the firm and requires manually pasting context, copying answers, and returning to one's own system to update manually; Referent, by contrast, works based on the firm's real-time case file context. For example, when asked "What needs my attention today?", it answers based on actual business and directly performs work—creating case files, drafting follow-ups pending approval, filing documents in the correct place, and setting reminders. A chatbot is an assistant that must be directed turn by turn; Referent is an operations layer that already knows the firm's state and can advance routine work on its own, while placing every client-facing action behind lawyer sign-off.
Limitations and status
Referent is currently in a private beta for boutique law firms, requires an application for access, and offers a personal onboarding call. The official disclosure states that the first beta cohort closed applications on August 7: 20 spots received applications from 450+ law firms.
This page does not provide specific pricing, supported languages, complete feature differences across platforms, or other such information; it is not yet available.
Other information
The official blog mentions topics including: practical workflows for Claude Code Mods for lawyers, the meaning and use of a legal AI harness, application data for the first cohort, and more. The founder and CEO is George Zaletski, officially described as having fifteen years of experience building technology, having worked at a law firm, and having four years in legal tech.
