Agents work the queue with you.
Five kinds of agent run alongside the firm — on the phone, in the pipeline, on every lead and matter. They answer, summarise, classify, fetch and explain. Lawyers still decide.
How we think about it
Capture → AI → human decide → explainable money. That’s the loop. Agents capture and process the work; people still decide; the financials behind every decision stay drillable to the row. The agents do not replace the lawyer — they take the admin out of being one.
The five agent types
01
Voice agent — answers, qualifies, books, creates the lead
Every inbound call is answered by a voice agent (ElevenLabs voice, Claude tool-calling). It captures the enquiry, asks the right qualifying questions for the practice area, books a consultation onto the right fee-earner's calendar, and creates the lead in the pipeline with all the context already attached. Per-firm config controls the script and voice — and the call is logged like any other.
02
Data agent — chat over the matter and finance data
Ask the firm's data in plain English — query contacts, matters, fees, balance and compliance status. Sessions are scoped (global, report, matter, contact) and RBAC-gated, so a fee earner sees only their matters and a billing clerk sees the firm's books. Persistent sessions hold up to fifty turns; the agent is read-only — it tells, it doesn't change.
03
Summary agents — keep every lead, matter and recording up to date
Lead summaries, matter summaries, contact summaries and recording summaries are generated by Claude Sonnet, cached on the record, and lazily refreshed when context changes (new transcript, new note, status change). Hover any lead or matter for the current summary; click refresh to regenerate.
04
Classification agents — relevance, quality, conflict
Every lead is scored by Claude Haiku — relevance to the firm's practice areas, narrative quality of the description, and a conflict-of-interest scan for risky keywords. The scores drive routing, gating and quarantine. Cheap, fast, on every record.
05
Activity feed — every move, visible
Every summary regeneration, every classification, every flag raised by an agent shows up in a realtime ops feed. Open views auto-refresh when an agent has done something relevant. No black box — what the agents do, the firm sees.
Models, scoping and safety
Specific by design.
Summaries and the data agent run on Claude Sonnet; classification runs on Claude Haiku; the voice agent uses ElevenLabs for voice and Claude for tool-calling. The model ID is centralised in one place so a deprecation never silently breaks the firm.
Every agent is scoped by RBAC: a fee earner only ever sees their own matters; a partner sees the firm read-only; a billing clerk has write on payments and recovery; an observer is read-only. The agents inherit those gates — they cannot see what the user cannot see.
The data agent is read-only by design. Voice agents create leads through a single, audited tool call. Nothing an agent does is invisible; everything lands in the activity feed.
Request access
We’re working with a small number of firms first.
Leave your work email and we’ll be in touch.