Counsel · After-hours intake, attended. - Vertical Agent Spec
One-line definition
An agent that answers after-hours calls, screens prospective clients, runs preliminary conflict checks, and deposits a structured intake summary plus draft retainer on the attorney's desk before the next business day.
The workflow it owns end-to-end
- Answers inbound calls to the firm's number between configured after-hours windows (typically 6 p.m. to 8 a.m.) using a warm, low-latency voice model trained on intake cadence for legal contexts.
- Conducts a structured intake interview: matter type, incident date, opposing parties and their attorneys, jurisdiction, urgency level, and preferred callback time.
- Queries the firm's existing client database (or a manually maintained adverse-party list) to surface potential conflicts before the call ends and flags anything that requires attorney review.
- Generates a draft retainer pre-populated with the caller's details, matter type, and the firm's standard fee structure, held in draft status pending attorney approval.
- Delivers a consolidated packet to the attorney by configured morning deadline: call transcript, structured intake form, conflict flag status, and draft retainer link.
What it knows that a generic LLM doesn't
- State-specific intake requirements: what fields are mandatory for a contingency agreement in California vs. a flat-fee criminal defense engagement in Texas.
- Conflict-check logic: how to parse "my husband's company" or "the other driver's insurer" into discrete adverse parties and cross-reference them against matter records.
- Matter-type triage vocabulary: the difference between a "slip and fall" that may be a premises liability case vs. a workers' comp case, and why that distinction changes the intake form entirely.
- Distress call management: callers who phone attorneys after hours are frequently in crisis. The agent knows when to slow down, repeat itself, and avoid dead-end prompts that cause hangups. It does not push toward upsell or schedule-optimization logic the way a generic voice bot would.
- Retainer format expectations: attorneys expect specific recitals, fee definitions, and signature blocks. The agent outputs a document that looks like a draft a paralegal produced, not a form letter.
- Bar-number and jurisdiction validation: the agent knows to ask which state the matter arises in and adjusts its language to avoid anything that reads as legal advice, which is the product's single largest liability surface.
What it explicitly declines
- Legal advice of any kind: the agent will not interpret law, predict case outcomes, advise on whether to sue, or tell a caller what their rights are.
- Final conflict clearance: the agent flags potential conflicts and surfaces the data; only the attorney clears or waives a conflict. The agent never tells a caller "you're cleared, we can take your case."
- Retainer execution: the draft document is never sent to the client for signature without attorney review and approval. The agent creates; the attorney authorizes.
- Emergency escalation outside the firm's protocol: if a caller indicates they are in physical danger or need emergency services, the agent instructs them to call 911 and ends its intake role there.
Tools and integrations required
- Voice infrastructure: Twilio for call routing and recording, with a low-latency voice model (Cartesia or ElevenLabs) for natural call cadence.
- Practice management: Clio or MyCase for existing client and matter records, used for conflict cross-reference and post-call intake deposit. Without this integration, the product cannot deliver on its core conflict-check promise, and manual re-entry will kill retention within 60 days.
- Document generation: a template engine (Documate or a custom layer) that merges intake data into jurisdiction-specific retainer templates maintained by the firm.
- Attorney inbox delivery: email with PDF attachment plus a web-based review link; optionally a Slack or Teams notification for firms that use those.
- Calendar: Calendly or Clio's built-in scheduling to offer callback slots the attorney has pre-approved, avoiding double-booking.
Trust escalation: when it pings a human
- Any call where the caller names an adverse party that matches an existing client record, even a partial match: the agent completes the intake, flags the conflict at high priority, and withholds the draft retainer until the attorney resolves it.
- Calls involving potential criminal matter where the caller indicates they are currently in custody: the agent captures basic information and immediately texts the attorney, because time sensitivity here is different from a standard intake.
- Any caller who expresses suicidal ideation, describes an ongoing domestic violence situation, or otherwise triggers a duty-to-report condition: the agent pauses intake, provides crisis line numbers, and sends the attorney an immediate alert.
- Any call where the caller's matter type falls outside the firm's configured practice areas: the agent completes intake but routes the packet to a referral queue rather than generating a retainer draft.
Pricing model
The honest pricing model here is per-completed intake, not a monthly seat fee. A rate of $35 to $55 per intake packet delivered and accepted by the attorney, with a monthly floor of $150 to cover availability costs during slow months, fits the economics. A solo practitioner converting two after-hours calls per week into retained clients at $1,500 average matter value will pay this readily. What they will not pay is a flat $300/month when call volume drops to two in a bad month. Per-intake pricing aligns the agent's incentive with the firm's outcome and removes the cancellation trigger that kills SaaS retention in low-volume professional services.
Differentiation from a generic LLM wrapper
A solo attorney who pastes their intake form into Claude gets a chatbot that cannot answer the phone, has no access to their matter records, does not know their retainer template, and has no mechanism for flagging a conflict against live firm data. Counsel's actual defensible position is not the AI model, which any competitor can replicate; it is the pre-built Clio and MyCase integrations, the jurisdiction-specific retainer template library, and the distress-caller voice tuning that reduces abandonment in the call types that matter most. Those three things together take roughly six months to build correctly and are genuinely difficult to copy at the margin. If the integrations ship incomplete or the template library is thin at launch, that differentiation collapses and the product is, in fact, just a wrapper.