BuddyLine — Voice-first AI companion for seniors

Field	Value
Status	Idea / strong candidate — not starting during the current focus window (created 2026-04-23)
Type	Consumer product / B2C2F (senior-to-family) — voice-first agent
Access surface	The senior's existing phone — inbound and outbound voice calls + SMS fallbacks; no app, no device, no download for the senior
Family surface	Lightweight web dashboard for setup (schedules, reminders, preferences, alert contacts)
Earliest revisit	2026-07-18 (end of current focus window)

One-line brief

A phone-number-based AI companion that seniors reach just by calling or being called — natural voice conversations, SMS fallbacks, identifies as AI immediately, no technology learning curve. Family members manage setup and oversight via a separate dashboard; the senior never has to install or configure anything.

Why this is on the wiki as a project and not just a note

Two reasons:

1. Unusually strong fit with the goals thesis. BuddyLine is a clean hit on all three legs — context depth (the per-senior model of routines, medication times, tone preferences, and topics they actually engage with), trusted access (the agent calls them and handles real daily scheduling under family-granted scope), and durable trust (the whole experience is explicitly designed around not breaking trust: identify as AI, stay out of regulated medical territory, never pretend to be human, never overpromise). It is the literal embodiment of the "agents with trusted access to your life" bet, applied to a demographic where the trust bar is high and the "just meet them where they are" posture is required rather than aesthetic.
2. Natural cross-project synergy with triviabalance. Both target seniors, both want to scale engagement content (trivia + balance exercises on one side; companionship + reminders on the other), and both benefit from the same context layer about what seniors actually stay engaged with. BuddyLine's real-world usage data is exactly the content signal TriviaBalance would want, and TriviaBalance's exercise + trivia library is a ready-made content source BuddyLine's calls can pull from. That synergy was previously unnamed; logging the project makes it explicit.

Core concept

A dead-simple, phone-number-based AI companion seniors reach by calling or receiving a call on their existing landline or cell phone. The experience should feel like talking to a friendly buddy, not using technology. All intelligence lives server-side so the senior surface is maximally low-friction — the senior does not install anything, does not sign in, does not learn an interface.

Family members get a separate, modest web dashboard for setup and oversight: schedule daily briefings, set up medication / appointment reminders, set preferred tone and topics, manage alert contacts.

Key features (MVP)

• Daily morning briefing call — weather, the senior's personalized schedule, gentle reminders.
• Medication + appointment reminders — both voice (outbound call) and SMS fallback when voice is inconvenient or missed.
• Voice-based calendaring — "Remind me to…" captured in conversation and surfaced back at the right time.
• Casual companionship chats — low-stakes, open-ended conversation, especially useful for loneliness. Explicitly positioned as a supplement, not a replacement for human contact.
• Simple family check-ins / status updates — senior can say "let Emma know I'm doing fine today" and the family dashboard receives it.
• One-command family alerts — if the senior flags distress or doesn't respond to a scheduled call, family gets notified according to preferences.

Technical approach (sketch)

• Senior interaction surface: phone calls (inbound + outbound) and SMS. No app, no download, no new device. Telephony via a programmable voice provider (Twilio / Telnyx / SignalWire class); exact choice TBD when the project is un-shelved.
• STT / TTS: xAI's low-cost voice models are the current bet (the same "voice pipeline got cheap enough to try seriously" shift that makes phone-first companionship plausible).
• Brain: Grok as the orchestrator (same family as the rest of the fleet). Per-senior context layer kept alongside the rest of the wiki/agent substrate.
• Scheduling: cron-style agentic reminders — same pattern the rest of the fleet already uses (see projects/crusty daily routines). Draft → (family-gate where appropriate) → execute → verify.
• Identity + privacy: per-senior account, family-scoped permissions, never mix seniors' data across accounts. Family members configure the dashboard under their own account; the senior identity is a managed sub-profile.
• Visual information fallback: anything visual (maps, recipes, a map to a doctor's office, a photo the family sent) is sent by SMS link and opens in the senior's native phone apps — BuddyLine never tries to be a UI.

Guardrails (things explicitly not to do)

• Never pretend to be human. Always identify as AI up front: "Hi, this is BuddyLine, your friendly AI companion…" Deepfake paranoia is real and well-earned in this demographic — honesty is also the cheapest trust-building move available.
• Stay out of regulated medical territory. No clinical advice, diagnoses, or HIPAA-heavy workflows. Companionship + light reminders only. If this ever pivots into medical, the compliance posture is a separate, very different project.
• Do not force seniors into apps or new interfaces. Zero learning curve is the entire point. Any feature that requires the senior to learn something new is a feature that doesn't belong in v1.
• Do not build physical hardware. A dedicated "BuddyLine phone" kills portability, ties the product to a hardware supply chain, and undoes the "uses the phone they already own" distribution advantage.
• Do not overpromise. Position it as a helpful supplement to real human contact, not a replacement — for both ethical reasons and to avoid the backlash when (not if) a family member finds the pitch patronizing.

Strategic moat

Early real-world usage data: which topics seniors actually stay engaged with, which reminder styles work best, which tones land, how speech patterns vary across cohorts, where frustration appears. That's context-rich demographic data that big labs can't easily replicate from pretraining alone — and it compounds per-senior over months. Same "context depth is the moat" bet the rest of the fleet is built on, applied to a demographic where the data is particularly scarce.

Where this fits in the fleet (see goals)

Possible framings, in rough order of commitment:

1. Concrete instance of Seat 5 (Consumer-product voice — conditional). Seat 5 on goals is currently a placeholder ("if a LunarCast-style consumer product resurfaces…"). BuddyLine could be the Seat 5 product rather than the abstract description — same "calm, no-BS, trust-first" consumer stance as projects/lunarcast, but on the voice surface instead of iOS.
2. Possible new Seat 7 (Senior companionship / voice) in its own right. Seat 5 was framed around a LunarCast-style stance, not specifically around a senior-care product; if BuddyLine goes live it may justify its own seat because the trust surface is materially different (voice, outbound calls, real-time conversation with a vulnerable demographic → higher blast radius than anything on the current roster, including Seat 4 Coaching).
3. Senior-audience substrate shared with triviabalance. Whether or not it gets its own seat, the content pipeline (what seniors stay engaged with, what reminders work) is a shared asset across BuddyLine and TriviaBalance. This alone is worth logging.

Decision on (1) vs (2) is deferred until at least the first prototype — there's no cost to leaving the seat ambiguous today.

Why this is not active during the current focus window (→ 2026-07-18)

The discipline reason, in order of weight:

• Focus-window integrity. The current window is Waypoints 1 + 2 (Aye Robot + Operator Agent). Starting BuddyLine now would still inflate the portfolio past the intentional two-waypoint frame.
• The thesis fit is durable, not time-sensitive. Voice-first senior companionship isn't a trend window — if anything, the longer the fleet's trust-architecture practice matures on Waypoints 1 + 2 first, the safer it is to aim that practice at a high-stakes voice surface later. Boring weekly pressure on a comic account and an EV / physical-operator node is the right pre-work.
• Real infrastructure pre-reqs. BuddyLine needs a telephony provider, cheap reliable STT + TTS (the xAI voice-model bet), a family-dashboard web surface, and per-senior account infra — none of which exists yet. Crusty model routing + xurl are no longer the block (shipped 2026-04-27 on projects/crusty); the wrong order of operations is still to build a full telephony + dashboard stack before Waypoints 1 + 2 have more reps on the Telegram-gate + ops pattern.

Revisit trigger: the default revisit is 2026-07-18 at the end of the current focus window. Earlier triggers that would pull the revisit forward: (a) a family member in Darcy's circle asks for a BuddyLine-shaped thing for a real senior (concrete pull > abstract roadmap); (b) projects/triviabalance starts generating senior-usage data that explicitly wants a conversational surface; (c) xAI's voice pipeline reaches a price/quality point that changes the math materially.

Open questions (to answer when un-shelved)

• Telephony provider. Twilio / Telnyx / SignalWire / other — driven by STT/TTS integration cost and per-minute pricing at the expected call cadence.
• Senior onboarding flow. Does the family member register from the dashboard and the system then calls the senior to introduce itself? Does the senior call in first? Does it ride on an existing number the senior already knows?
• Latency tolerance. Voice conversations tolerate much less latency than async chat. Budget per turn matters; will affect STT/TTS choice more than it will affect brain choice.
• Call-screening reality. Many seniors route unknown numbers to voicemail; caller-ID and call-from-consistent-number becomes a product feature, not a detail.
• Monetization model. Family-paid subscription is the obvious default; the exact price point and what's included vs add-on is open.
• Cross-project content pipeline with triviabalance. Shared library? Shared per-senior profile? Separate products with a data-sharing agreement? Pick once both are real enough to matter.

Related

• What-Im-Working-On — lives under Context / other lines during the current focus window.
• goals — maps cleanly to the thesis (context depth + trusted access + durable trust); candidate concrete instance of Seat 5 or new Seat 7.
• projects/triviabalance — shared senior audience; shared candidate content pipeline; mutually reinforcing moats.
• projects/crusty — same orchestration pattern (Grok brain, draft → gate → execute → verify, per-owner context layer); BuddyLine would be a second instance of that pattern against a consumer audience.
• projects/lunarcast — reference for the trust-first consumer stance (CloudKit no-sign-in, no ads, no algorithmic push). BuddyLine's version of that stance is identify as AI, no medical overreach, phone-first, supplement not replacement.

Raw sources

• raw/buddyline-summary.md — verbatim 2026-04-23 ingest.

Meta

• Created: 2026-04-23 from a same-day message capturing the concept + explicit guardrails + a parallel voice-for-education note.
• Current posture: Logged as a strong strategic candidate. Not active during the current focus window; revisit at 2026-07-18 unless a concrete pull arrives sooner.
• Next review: 2026-07-18 with the rest of the quarterly goals review.