The calls your office takes all day, answered carefully.
VoiceRobot answers inbound calls and texts for law offices. It verifies who is calling, reads from your systems rather than guessing, and hands to a person the moment the answer stops being one it is allowed to give.
Inbound voice and SMS run on one compiled runtime today — a caller can be verified, hear what a record says, be booked in, get a text confirmation, and reach a person. The intake playbooks for specific practice areas are being written now, and outbound reminder campaigns await carrier registration (A2P 10DLC). We would rather tell you which half is which than demo something you cannot buy yet.
What it handles
One reception agent answers and works out why they called, then hands the conversation to the specialist that owns it — with verified identity and a typed summary, never the raw transcript.
Answering the line
Greets every caller with the recording disclosure your state requires, finds out why they are ringing, and takes a name and a number before anything else.
Existing clients
Verifies identity against two factors before reading anything from a record aloud, then answers from your systems rather than from a model's memory.
Scheduling
Offers times inside your staffed hours, reads the appointment back before it is written, and confirms by text.
After-hours emergencies
Evaluated before the after-hours branch, so an arrest at 2am reaches on-call instead of voicemail.
Reaching a person
Transfers inside staffed hours. Outside them, or when nobody picks up, it files a callback and says so — it cannot promise a call it has not booked.
Anything it should not decide
It gives no legal advice, never says whether the firm will take a matter, and quotes no fee a schedule does not carry. Those refusals are enforced in code, not asked of the model.
The parts a firm's ethics counsel asks about
Compliance is code, not instructions
A prompt is guidance; an executor check is a control. The identity gate, the calling window, the consent check, and the approval queue for anything outside policy are enforced where the decision happens, and a flow that lacks one cannot be published.
It says when it cannot know
When your records system is unreachable the agent acknowledges the limit and offers a callback. It does not guess. That path is a required edge in every flow: a graph without one does not compile.
Disclosures are spoken, not paraphrased
Recording disclosures and policy statements are literal text handed to speech synthesis, never routed through a model. In a two-party-consent state, a flow whose greeting lacks one cannot be released.
Your data stays yours
Every tenant-scoped table carries row-level security forced at the database, and the application connects as a role that cannot bypass it. Isolation is a property of the database, not a filter someone remembered to write.
Your staff author the calls
Conversations are built as a flow — steps and the paths between them — in a visual builder, by the people who know your policies. Every change is compiled, validated, and released to a QA environment first; promoting it to production ships the exact artefact that was tested, and rolling back is repointing at an earlier one.
Nothing that fails validation can be released. A step that reads from your records without a path for the system being down is refused, because that missing path is a dropped call at the worst possible moment.