Product Roadmap
Current State (2026-02-26)
CareSupport runs as a single-family pilot (Kano family). One care recipient (Degitu), one registered member (Liban), SMS via Linq/iMessage. Files are the database. No self-service onboarding. No multi-family routing beyond directory scanning.What's Proven
- 13-step SMS pipeline works end-to-end
- Approval gating blocks sensitive changes until confirmed
- PHI filtering, role-based access, audit logging all mechanical
- Family.md + member profiles update through conversation
- Outreach queuing (needs_outreach → Linq create_chat) works
What's Broken
- Agent asks excessive clarifying questions instead of acting on clear input
- 6 family members listed with phone numbers → 0 added to routing.json
- Pending approval for care recipient update → user hasn’t been prompted to confirm
- No per-family learning — corrections are global only
- Relationship context stored as single field (to care recipient), not member-to-member
Implementation Strategy: Land → Pro → Agency
The Expansion Sequence: Land with Family → pull in Pro via invites → bring in Agency with shared-timeline proof. Seed each step with clear wins.Phase 2: Pull in Pro
Enable a professional to coordinate multi-family work while plugging into the family’s Circle
Phase 0: Fix the Pilot (Now)
Goal
Make the Kano family work properly before scaling anything.Conversation Skills
What: Add
skill.md — social skills that guide conversation flowWhy: Agent should prompt “Want me to invite them?” after receiving member list, not ask 3 more clarifying questionsContext Prioritization
What: Define mandatory vs optional fields in family.md and member.mdWhy: Agent wastes turns asking for blood type when it should be saving the 6 phone numbers it already has
Per-Family Lessons
What: Add
families/{id}/lessons.md — local correctionsWhy: “Degitu prefers Auntie” shouldn’t leak to other families; “always confirm before adding members” should be globalResolve Pending State
What: Clear the pending approval backlog, add the 6 family membersWhy: The pilot is stuck because the agent over-gated a straightforward update
Phase 1: Land with Family (Rob’s Network)
Timeline: 0–30 Days
Goal: Prove NHS lift in a real Family Network (Rob + Marta), then expand Pro → Agency with clear wins.Setup
- Create Family Circle
- Set Policy Pack
Family.Pro-Integrated - Set coverage window
- Import calendars
- Invite close supporters
Quick Wins (≤7 days)
Baseline Metrics
Baseline Network Health Score (NHS) auto-generated
First Gap Alert
First gap alert detected and resolved
First Handoff
Handoff summary captured at least once
Coverage Improvement
Coverage % +10pp vs. baseline OR gap minutes/week −20%
Features to Emphasize
- Today/Timeline view
- Task management
- Schedule conflict detection
- Availability rules
- Handoff summaries
Complete Single-Family Depth
| Item | Success Metric |
|---|---|
| All members registered | 7/7 people in routing.json with profiles; all can text CareSupport and get context-aware responses |
| Transportation schedule | Mon–Fri pickup/dropoff in family.md This Week section; schedule populated, visible to drivers |
| First outreach | CareSupport texts Solan or Yada about a shift; outreach sent, reply received, logged |
| First handoff | Driver change captured and next driver notified; handoff summary in timeline |
| NHS baseline | Network Health Score v0 calculated; Coverage %, gap minutes, time-to-fill measurable |
Trigger to Phase 2: Uncovered hours >15% OR recurring tasks require pro skills → Invite a Pro nudge.
Phase 2: Pull in Pro via Invites (CareGiver OS)
Timeline: 31–60 Days
Goal: Enable a professional to coordinate multi-family work while plugging into the family’s Circle.Flow
Quick Wins (≤7 days)
First Session Logged
First session logged & summarized; family sees it in Timeline
Fast Gap Fill
Time-to-fill for next gap < 24h
Adherence Improvement
Adherence improvement for at least one med/appointment
First Payout
Pro issues first invoice/payout
Incentives
- 14-day Pro trial
- Referral credit
- “Verified Pro Profile”
Multi-Family Architecture
Make the system capable of running 2+ families without operator intervention.| Item | Challenge |
|---|---|
| Family creation CLI | python scripts/create_family.py --name "tefera" --coordinator "+1..." — Seed routing.json, family.md, member profile from templates |
| SMS self-service signup | New number texts in → “Start a care network” flow — Need to distinguish new family vs existing member on unknown number |
| Cross-family isolation | One member in 2 families (e.g., caregiver serves multiple) — routing.json currently maps phone → one family. Need multi-family resolution |
| Global lesson graduation | Local lesson appears in 2+ families → promote to global — Need a review mechanism, not automatic (family-specific context could be wrong globally) |
Trigger to Phase 3: Pro/Family references an agency OR backup staffing need emerges → Share Timeline with Agency prompt.
Phase 3: Bring in Agency via Shared-Timeline Proof
Timeline: 61–90 Days
Goal: Demonstrate value to the agency without heavy integration.Flow
Quick Wins (≤14 days)
Reduced Calls
Coordinator calls/emails reduced for that family
Better Coverage
Coverage % up; support tickets down
NHS Improvement
Agency sees NHS lift and handoff quality metrics
Next Steps
Offer deeper adapters (HR/payroll/EVV) based on demonstrated ROI
Pro Integration Features
| Feature | Description |
|---|---|
| Pro invite flow | Family coordinator invites pro → pro joins with scoped access |
| Multi-family dashboard | Pro sees schedule across families they serve |
| Session logging | Pro logs visits, family sees summaries in timeline |
| Availability rules | Pro sets hours, system prevents invalid assignments |
Exit Criteria (90 Days)
Family NHS
NHS ≥ +10 over baseline and sustained 2+ weeks
Pro Retention
30-day retention ≥ 70%; ≥ 80% sessions with handoffs
Agency Impact
Pilot demonstrates measurable reductions in calls/tickets
KPI Ladder by Stage
Phase 1: Family
- NHS Delta: Improvement over baseline
- Gap minutes ↓: Reduction in uncovered time per week
- Time-to-first-resolution: Speed of gap closure
- # helpers activated: Number of team members actively using system
Phase 2: Pro
- Time-to-first-session: Speed from invite to first logged session
- % visits with handoffs: Quality of care continuity
- Pro 30-day retention: Stickiness of CareGiver OS
- Invoice cycle time: Business operations efficiency
Phase 3: Agency
- Linked clients count: Number of families connected
- Support contacts ↓: Reduction in coordinator burden
- Coverage % ↑: Improvement in shift coverage
- NPS lift: Partner satisfaction improvement
Scaling Challenges
Challenge 1: Scaling Families
Problem & Constraints
Problem & Constraints
Problem: Every new family requires manual directory creation, routing.json seeding, family.md templating, first-message-via-Linq to establish chat_id. This is 6 manual steps that require operator knowledge.Why it’s hard:
- Phone number is the primary key, but one person can be in multiple families
- Chat_id (Linq’s UUID) is only known after first message — chicken-and-egg
- Template family.md has sections that may not apply to every family (not everyone tracks medications)
- No rollback if onboarding fails halfway
- Must work over SMS (no web UI for onboarding)
- Must preserve file-as-database architecture (no external DB)
- Must maintain approval gating for sensitive data from day one
Proposed Approach
Proposed Approach
- CLI tool first (operator-assisted, Phase 2)
- SMS self-service second (fully automated, Phase 2 stretch)
- Web dashboard third (Phase 3+, only if needed)
Challenge 2: Context Explosion
Problem & Approach
Problem & Approach
Problem: As families grow, the amount of context loaded per message grows. family.md Current section has a 2000-token soft limit, but with 7+ members, active medications, and weekly schedules, it’ll exceed that.Why it’s hard:
- Every message loads Current section in full
- Agent needs enough context to be useful but not so much it hallucinates
- Different messages need different context (schedule question vs medication question)
Challenge 3: Conversation Quality at Scale
Problem & Approach
Problem & Approach
Problem: CareSupport’s conversation quality depends on prompt engineering in SOUL.md + capabilities.md. As we add families with different dynamics, a single prompt may not generalize.Why it’s hard:
- Each family has different communication norms
- Some families are terse, some are verbose
- Cultural context matters (naming conventions, family structure, communication expectations)
- Global lessons from one family may not apply to another
skill.md for conversation patterns + per-family lessons.md for corrections. Global lessons only for universal patterns.Network Health Metrics
North Star: Network Health Score (NHS)
Definition: Single score per network to prove coordination valueCoverage %
Percentage of target coverage window actually covered
Gap Minutes
Total uncovered minutes per week
Time-to-Fill
Median time from gap detection to resolution
Adherence
Medication and appointment adherence rates
Handoff Quality
Percentage of shifts with proper handoff documentation
Coordination Time
Coordinator time saved on scheduling and follow-ups
Product KPIs
- NHS delta over baseline: Primary success metric
- Network NPS (coordination lead view): Satisfaction measurement
- CareGiver OS retention: Professional stickiness
- Partner retention (platform/agency): B2B success
Technology Roadmap
Now: Lighthouse Pilot
- Coverage workspace
- Gap/conflict detection
- Handoffs
- Availability rules
- Decision trace capture
Next: Multi-Family Operations
- Ranked proposals
- Fairness heatmaps
- Opt-in automations
- Precedent search UI
Later: Advanced Capabilities
- External calendar & wearable signals
- Proactive “coverage health” alerts
- Community routing
- Simulation queries
Risks & Mitigations
| Risk | Why It Matters | Mitigation |
|---|---|---|
| Cold start in multi-sided network | Need families and pros and partners | Stage GTM; seed Family Hubs; recruit pros via value props; pilot partners |
| Integration friction | Agencies/marketplaces are busy, legacy stacks | Provide adapters, playbooks, commercial incentives (retention/NPS lift) |
| Compliance variability by state | Product behavior must adapt | Policy graph + jurisdiction packs; local counsel & templates |
| Perception as competitor | Partners may fear overlap | Strict neutral posture; published interoperability charter; white-label options |
| Data trust & privacy | Must be trusted by families and pros | Transparent controls, consent receipts, minimal data exposure by default |
Strategic FAQ
How are you different?
How are you different?
We equip each care network with network-specific infrastructure (Circle + Policy Pack) and prove value via a Network Health Score. We expand Land → Pro → Agency with clear wins at each step.
Are you a marketplace?
Are you a marketplace?
No. We’re the coordination layer used after the match and during the relationship. Families, independent pros, agencies, and platforms each get tools tuned to their network.
Why CareGiver OS?
Why CareGiver OS?
It elevates independent caregivers into technology-enabled professionals who can run multi-family practices, while linking cleanly into family circles and agency views.
What's the business model?
What's the business model?
B2C (Family), B2Pro (CareGiver OS), B2B (Agency), plus Platform APIs. All mapped to demonstrated NHS improvements.
Why will networks adopt this?
Why will networks adopt this?
Measurable NHS lift (fewer gaps, faster fills, better handoffs), reduced coordination burden, and simple adapters (CSV/ICS/API-lite) with a path to deeper integrations.