What it does, in detail

Structured extraction

Every inbound email is parsed by Claude with a strict JSON tool schema. We extract:

• Insurance carrier, member ID, plan type, group number, self-pay flag
• Presenting concerns (verbatim and a one-line clinical summary)
• Prior therapy history
• Modality preference (in-person / telehealth / either)
• Availability — preferred days, times of day, specific windows
• Crisis flag and rationale

Crisis routing

If the email contains active suicidal ideation, plans, means, or imminent self-harm signals, we:

1. Send the client a same-second auto-reply listing 988, 911, and 741741.
2. Mark the intake as urgent in your dashboard.
3. Page you in real time, separate from the daily digest.

Crisis detection runs as a deterministic regex layer plus an LLM signal — both must clear before we de-escalate. False positives are tolerable; false negatives are not.

Smart follow-ups

When insurance or availability is missing, we ask the client one polite follow-up question — not a five-question form. When they reply, the new message is threaded back to the same intake record.

Calendar-aware slot proposals

Optionally connect your Google Calendar (read-only). For each new intake, we propose three real open slots in the auto-reply that match the client's stated availability window. They reply with a pick; you confirm.

Daily digest

At 6pm in your timezone (configurable), every routine intake from the day arrives in a single email. Crisis cases are at the top, followed by routine inquiries. Each row is a clean, structured summary — never the verbatim email body.

Dashboard (optional)

If you want to see everything in one place, your dashboard at /inbox lists every intake with full search, filters, and an audit-logged "view" event. Most users check it once a week.