ElevenLabs x Sauna Hack Night

# 👻 Ghostmeet Video calls where the people who couldn't make it still get a say. If an expected\ guest doesn't show up but left some context, their **voice proxy** joins: it answers\ for them in their own voice (ElevenLabs) when named, and never makes things up. **Live demo:** <https://ghostmeet-vyezdhds.sauna.new/?room=DEMO01> ## What it does - **Voice in, voice out.** Tap 🎤 Speak, say your line, and the named guest's proxy\ answers out loud (speech-to-text + LLM + text-to-speech, all via ElevenLabs + Sauna). - **Grounded, honest proxies.** Each proxy answers only from the context it was given.\ If it doesn't know, it says so instead of inventing. - **Calendar-aware.** For availability questions, a proxy checks the connected Google\ Calendar; when there's agreement, the organizer schedules the meeting in one click. - **Wrap-up.** The organizer ends the call and emails a summary (SUMMARY / DECISIONS /\ PENDING ACTIONS) plus what each proxy said, so absent people can confirm. - **Multi-device.** Join from a laptop or phone with a link; a TURN relay keeps\ cross-network (wifi ↔ cellular) video connected. ## How to use 1. **Organizer:** open the app, create a room, list the expected guests, and give each\ one a context note (+ optional voice, email, and calendar). 2. Share the `?room=CODE` link. Guests tap **Join call** and allow camera/mic. 3. A guest who doesn't show up (and has context) becomes a **proxy** — name them in the\ chat or with 🎤 Speak to hear them reply. 4. Schedule the follow-up and send the summary from the organizer panel. ## How it's built Sauna App — WebRTC mesh + WebSocket signaling, an app-owned SQLite database (rooms,\ guests, transcript, presence), and Sauna's LLM + ElevenLabs endpoints called from the\ handler. Google Calendar and Gmail are the organizer's connected accounts. ## Honest limits - The proxy voice plays as audio in the call, not injected into the raw WebRTC stream. - Proxies are **AI personas** built from the context you provide — not real Sauna agents.\ The peer-to-peer path (pulling a real person's context from their shared Sauna) is\ wired but inactive until a teammate shares their Sauna. - A reply takes \~6–8s (mostly model latency); a "thinking…" indicator shows meanwhile. - Calendar/email use the organizer's Google account; third-party availability needs\ those people to connect their own.

Sauna by WordwareElevenLabs

Why? I run 10 of coding sessions daily. And orchestration of them a headache I would like to control them over voice So I could listen to agent when in gym or driving How? Expose opencode rest api (https://opencode.ai/docs/server/) that runs on dev vm over tailsaced funel Connect to the voice chat app on sauna

Sauna by WordwareElevenLabsopencode, linux vm, systemd, tailscaled

PrepSauna

Its is a mock interview app made using Sauna, powered by 11labs for interviewer persona.

Sauna by WordwareElevenLabs

A voice-first memory layer for teams. Speak to Echo and it transcribes what you said, pulls out the facts, tasks, and decisions buried in the conversation, and keeps them as persistent context. Ask it something later and it answers out loud with that history in mind — so the things a team says out loud stop evaporating the moment the meeting ends. ElevenLabs handles voice in and voice out. Sauna carries the memory: the context Echo captures lands in the same AI-coworker layer that already knows the team's tools and habits, so spoken context becomes something the team can act on rather than a transcript nobody reads.

Sauna by WordwareElevenLabsPython, Gradio

A Chief Advisor that speaks first, remembers you, and shows its receipts. It reads your Gmail, Calendar, and Notes via Sauna, opens the conversation about the one decision that matters today, and talks it through with you by voice via ElevenLabs.

Sauna by WordwareElevenLabs

Voice-first AI work OS — reads your inbox, files Jira tickets, creates Google Docs and Excel sheets, and sends Slack messages while you drive. No screen. No typing. Just your voice.

Sauna by WordwareElevenLabs

Whentor Chief Adviser

Whentor AI Chief Advisor is a personalized AI that understands your life over time, follows up on what matters, and helps you make better decisions through text, voice, and calls.

Sauna by WordwareElevenLabs

Ctrl + Alt + Life

Ctrl + Alt + Life is an AI life-transition companion for people navigating layoffs, burnout, career changes, relocation, and other major moments of uncertainty. Instead of immediately pushing users toward resumes and job applications, the app begins with one human question: “Life happened. What happened?” It asks how the user is feeling, understands their current situation, and helps them decide what deserves their attention next. The app creates a personalized plan with practical priorities, emotional support options, local networking recommendations, and follow-up actions. Sauna powers the persistent profile, contextual memory, personalized planning, and proactive follow-up experience. ElevenLabs adds a natural voice layer so users can speak through difficult transitions rather than relying only on forms and text. The core idea is simple: computers have Ctrl + Alt + Delete when they crash. Humans deserve Ctrl + Alt + Life when life changes.

Sauna by WordwareElevenLabs

# Kin — an end-to-end voice agent for vulnerable populations including the elderly and disabled **Live demo:** <https://kin-mzuzhhlk.sauna.new/> A vulnerable person just talks. Kin knows them, acts for them, and loops in their guardian before anything risky happens. We demo with an 82-year-old, but the architecture fits anyone who needs a trusted person in their corner: elderly, visually impaired, motor or cognitive conditions, or anyone vulnerable to scams. Almost all consumer AI assumes a fluent, able, tech-comfortable user. Roughly 1 in 6 people worldwide live with a significant disability, and the population is ageing fast. Kin is built for the rest. ## What it does - **Talk naturally.** One big button, or hands-free "Hey Kin". Speak, and Kin answers out loud. - **Remembers.** Conditions, medications, contacts, and anything learned along the way ("mentioned chest pain, wants to book a doctor"). - **Acts safely.** Small everyday spending happens on its own. Larger or unusual requests pause and go to the guardian for approval, who replies with one tap. - **Screens for harm.** A stranger pressuring for money (roof repairs, fees, prizes) triggers the same guardian check, catching scams before money moves. ## How it works "ElevenLabs is the ears and mouth. The Sauna app is the memory and hands." - **Voice:** ElevenLabs speech-to-text and text-to-speech, called from the app backend. Warm, patient voice, captions on screen. - **Brain + memory + hands:** a Sauna App holding the person's profile and learned facts in its own database, deciding each turn with an LLM, and executing real actions. - **Guardian loop:** an over-limit request emails the guardian a real Approve / Decline link. Their tap flips the request live, and Kin tells the person the outcome. Everything runs in one deployed app. No separate servers, no glue code. ## Accessibility - Huge tap target, tap-once-and-talk (auto-stops when you finish speaking). - Adjustable text size and a high-contrast mode, both remembered between visits. - Replies shown as large captions and read aloud, so it works for low vision and hard of hearing. - The caregiver dashboard (memory, spending, settings) lives on a separate tab, so the person only ever sees the big button. ## 90-second demo 1. "How am I doing this week?" Kin recalls the refill due Friday and the dizziness on Tuesday. 2. "Send ten pounds to my granddaughter Emily." Done instantly, within her everyday limit. 3. "I need two hundred pounds for a man who called about my roof." Kin pauses, asks the guardian, the guardian's phone pings, they decline, and Kin tells her it is handled. ## Generalising The profile drives everything. Swap the 82-year-old for a visually impaired 30-year-old or someone with a learning disability: different voice pacing, different approval thresholds, same system. Personalisation is config, not code. ## Stack ElevenLabs (voice) + Sauna Apps (persistent backend, database, LLM, Gmail delivery). Guardian alerts over email today, with iMessage and WhatsApp as drop-in channels.

Sauna by WordwareElevenLabs

awesome voice agent in sauna house

a sauna house around animal farm where a pig is looking for food, ask the eleven lab voice agent for helping out

Sauna by WordwareElevenLabs

Vamos

EchoCheck is a real-time decision and commitment checker for teams. It listens during a conversation, detects when someone is about to make a promise or decision, and checks that statement against the team’s existing knowledge in Sauna. Instead of summarizing a meeting afterward, EchoCheck tries to prevent mistakes before the commitment is made.

Sauna by WordwareElevenLabs

A personalized ElevenLabs voice agent clone of me that my mother can call for IT assistance...instead of calling me. - ElevenLabs Instant Voice Clone (created with a reading of the Love Song of J. Alfred Prufrock on a DJI Mic) - Twilio Individual Phone Line API - BraveSearch MCP for User Manual Search - Calibrated Q+A with Bayesian Softening for User-Specific Instruction Adjustment - All created with tender digital love, togetherness, and Finnish Hospitality in Sauna AI

Sauna by WordwareElevenLabs

conversational animated sports companion

Sauna by WordwareElevenLabs

I'm a CS student at Berkeley, and like a lot of gen-z "devs", I learned to code by vibing — prompting an AI until something works, without always understanding why. It's fast, and it feels like magic, right up until it crashes. Then I'm stuck, because I never actually learned the thing that broke. DuckTape is my attempt to fix that. It's an AI coding buddy that talks to you like a real rubber duck — the old programmer's trick of explaining your problem out loud to a duck on your desk until the answer clicks. Except this duck actually talks back, and it remembers. It remembers what you're building, what broke last time, how you like to work. The goal isn't just to spit out fixes. It's to bring back the part of coding that vibing has quietly erased — the understanding, the back-and-forth, the "oh, THAT'S why it broke" moment. If you actually wanna learn coding + understand what ur building, get ducktape! A quick list of features: 🎓 Teaches, doesn't just fix — closes the vibe-coding understanding gap 👉 Point & explain — point at any code, it explains it out loud 💻 Web version + terminal — a CLI with the same brain and memory as the web GUI 🎙️ Talks with you out loud 🤖 Agent - can build new stuff & debug 🧠 Never forgets — remembers your project, stack, bugs, and fixes 🔗 One brain, everywhere — web, voice, and terminal share one memory

Sauna by WordwareElevenLabsReact + Tailwind, and a walking-duck mascot because obviously.

Daming

Fluent Me An AI English tutor. You chat with Kai (your tutor) — and every correction plays back in your own cloned voice, never Kai's. You literally hear the fluent version of yourself. Built for ElevenLabs × Sauna Hack Night (2026-07-16, SF). Hold to talk. A live waveform shows Kai hearing you as you speak. Every mistake becomes a memory card, and Kai hides the review inside normal conversation — no flashcards. Then you echo your fluent line back and get scored against yourself, five seconds ago. What it does Your voice, with fluent and native speech. Honest scoring (4 dims) — grammar + vocab (Claude), fluency (pace / pauses / fillers), and pronunciation as an ASR-confidence proxy, labeled as such Live waveform — real-time scrolling mic waveform while you speak, so you know it's listening Interview prep — mock interviewer in character + STAR feedback, closed with your own answer polished in your voice Presentation rehearsal — clarity coaching, per-section pacing, skeptical-audience Q&A Conversational spaced repetition — FSRS-style scheduling hidden in the chat, your life as the curriculum

Sauna by WordwareElevenLabs

zoos

Zootopic is a zoo narration and conversation AI bot that speaks with visitors about animal’s current activities, as well as keeping long term memory for the future. The bot keeps long term memory, recent context, and current activities in its memories. It uses ElevenLabs to speak with zoo guests, in an actual conversation.

Sauna by WordwareElevenLabs

Dead poet squad

A tool that taps into my linkedin to recommend 1 person each day i can network/reach out to

Sauna by Wordware

Digital system to prototype a custom 3d printed storage box

Sauna by Wordware

Sauna Match

We made Tinder for your Sauna agents. Uses Elevenlabs TTS and STT and the inbuilt sauna OpenAI exposable model as the LLM. Given each user's sauna has a lot of context about their user, the Sauna agents can literally have pre-qualifying conversations in the background asking each other questions about the opposing user until they both decide at the end whether to swipe right or swipe left. If they both swipe right, then a chat room for the actual users is set on the app.

Sauna by WordwareElevenLabs

hypeee

hypeee alarm to wake you up based on your calander

Sauna by WordwareElevenLabsgooogle calander

Taskly

# Taskly PRD (Product Requirements Document) ## 1) Product Overview Taskly is a local-first productivity app that turns natural-language tasks into scheduled Apple Calendar events, with conflict-aware reliability and human-friendly fallback suggestions. ## 2) Problem Statement Users need one place to capture tasks quickly and have them scheduled intelligently without calendar collisions. Existing workflows are fragmented across notes, reminders, and calendar. ## 3) Goals - Capture tasks in natural language with minimal friction. - Auto-schedule into available work windows. - Keep scheduling reliable even with real-time calendar changes. - Support practical constraints (e.g., mornings-only, weekdays-only). - Allow easy rescheduling/deletion directly from UI. ## 4) Non-Goals (Current Scope) - Multi-user collaboration - Cloud sync/accounts - Mobile-native app - AI-generated long planning narratives ## 5) Primary Users - Busy professionals managing personal + work task load. - Builders/creators with dynamic calendars and frequent context switching. ## 6) Core User Stories 1. As a user, I can paste multiple tasks in plain text and schedule all at once. 2. As a user, I can have Taskly detect duration and urgency hints. 3. As a user, I can avoid scheduling conflicts even if calendar changes right before save. 4. As a user, when conflicts happen, I get immediate fallback slot options. 5. As a user, I can edit, snooze, mark done, or delete events from the UI. 6. As a user, I can constrain scheduling to mornings and/or weekdays. ## 7) Functional Requirements ### 7.1 Task Capture & Parsing - Multi-line text input (one task per line). - Parse durations: `30 min`, `1 hour`, etc. - Parse intent keywords: `today`, `tomorrow`, `urgent`, etc. - Parse constraint keywords: `mornings-only`, `weekdays-only`, variants. ### 7.2 Scheduling Engine - Default work window: 09:00–18:00. - Search horizon: up to 21 days. - Slot resolution: 15-minute granularity. - Constraint-aware windows: - Mornings-only → 09:00–12:00 - Weekdays-only → Mon–Fri only ### 7.3 Reliability (Phase 2) - Final conflict-safe revalidation before writing new events. - Final conflict-safe revalidation before event move/update. - Structured conflict handling with fallback suggestions. - Fallback suggestions: top 3 valid slots over next 3 days. ### 7.4 Calendar Integration - Apple Calendar integration via AppleScript (`osascript`). - Two-way visibility in calendar view. - Edit event time and delete event by UID. - Dedicated calendar support (`CALENDAR_NAME`). ### 7.5 UI/UX - Premium dark dashboard visual style. - Today View (Now / Next / Later). - Reminder filters (All / Active / Done / Missed). - Undo support for done/snooze state changes. - Month/Week calendar view. - Constraint toggles in form: - Mornings only - Weekdays only - Results table includes applied constraints per task. ## 8) Non-Functional Requirements - Fast interaction on local machine. - Fail gracefully on Calendar permission issues. - Deterministic scheduling behavior. - Clear user feedback on failures/conflicts. ## 9) Tech Stack - Backend: FastAPI (Python) - Frontend: Jinja templates + vanilla JS + CSS - Integration: Apple Calendar via AppleScript - Runtime: local macOS app server ## 10) Current Delivery Status ### Phase 1 (Completed) 1. Task state display 2. Done/Snooze actions 3. Today View (Now/Next/Later) ### Phase 1.5 (Completed) - Reminder filters (All/Active/Done/Missed) - Undo for Done/Snooze ### Phase 2 Reliability (Completed) 1. Conflict-safe revalidation before writing events 2. Fallback slot suggestions on conflict ### Phase 3 Constraint Scheduling (Completed) - Keyword-based constraint parsing (morning/weekdays) - UI toggles for mornings-only/weekdays-only - Constraint-aware slot search and fallback suggestions - Constraints shown in scheduling results ## 11) Success Metrics - Scheduling success rate (without manual edits) - Conflict recovery rate (conflict resolved via fallback) - Duplicate/conflict incident rate - Average time from input to scheduled event ## 12) Risks & Mitigations - **Calendar permission issues** → clear setup guidance and error messaging. - **AppleScript fragility** → defensive parsing + fail-safe responses. - **Race conditions** → revalidation before write + fallback suggestions. ## 13) Roadmap (Next) - Smarter fallback ranking (priority, context, preference scoring) - Constraint presets (deep-work blocks, no-meeting windows) - Better event editor UX (title + duration edits) - Duplicate-detection rules - Weekly planning assistant mode ## 14) Open Questions - Should constraints persist as user defaults? - Should fallback suggestions auto-apply if user chooses one click action? - Should we support personal/work calendar filtering during scheduling (not only view)?

Sauna by WordwareElevenLabs

80% of the workforce is deskless. HVAC and mechanical techs can’t type on a job site, so site history lives in people’s heads and documentation never happens. A repeat truck roll costs $150–$500. Incumbents are forms-on-a-phone — built for dispatchers, hated by techs. What it does Recall — arrive at a site, ask by voice, get the full visit history, warranty, and access notes. Capture — ramble a messy post-job debrief; parts, follow-ups, and access codes land in shared memory. Multiplayer — the next tech (or Sauna at HQ) asks what happened — and hears what you just logged. How it’s wired ElevenLabs is the mouth and ears (Conversational AI + server tools). Sauna is the brain and shared memory (custom MCP connector). FieldMemory is the voice front door between the rooftop and the shop. ElevenLabs Voice agent Sauna MCP Team memory Claude Structure + brief SQLite Local memory layer Who it’s for Mechanical / HVAC / electrical contractors with 20–200 techs. Roughly $60–100 per tech / month. One avoided repeat visit per tech per month pays for it — and every debrief raises switching cost.

Sauna by WordwareElevenLabs

Meet the Risk Manager Agent — built on the Pillar Test™

Sauna by WordwareElevenLabs

A voice-driven design brainstorming whiteboard. Transcribes your voice (ElevenLabs) and distills speech into structured design memory (product, features, constraints, rejected ideas, open questions). Forks a new branch every time you say "or..." so alternative directions become their own visual timeline you can switch between. Remembers the whole session, so you can ask "what were the three constraints?" or "bring back the folding-arm idea" and it recalls instantly

Sauna by WordwareElevenLabs

My Grandma hates UI. She doesn't want to tap a button to ask a question. She wants to walk in the room and have the right thing happen — the weather, the prayer time, the call from her son, the reminder about her medicine, the photo from last week. You don't open an app. The app opens for you. The interface is the agent's judgment, not its controls. The UI is gone. The intelligence is in the back. It watches what you're doing, knows what you usually want in that moment, and surfaces it — mostly through voice, sometimes as a small floating bubble around the orb — and only when it actually helps. That's the shape of what's being built here. Argus is the surface area of a personal AI that doesn't have a screen. It runs in your browser, asks for camera + mic, and then disappears into the background. Every few seconds it captures one frame and a short audio clip from your environment, runs them through a vision model, and writes a structured memory of what's happening — what you're doing, where you are, who's around you, what it sounds like, how it feels. That memory grows. Over time it knows your patterns, your people, your places, your moods. When something genuinely matters, it speaks. Otherwise it stays quiet. The only thing on screen is an orb. A black background, a glowing blob in the middle that breathes when it's idle, pulses when it's listening, spins when it's thinking, and bounces when it's talking. Tap it to begin. Tap it again to ask something. Tap it again to stop. Everything else — the camera preview, the timeline, the concept graph, the typed question box — lives behind a small "Old" button in the corner, for the days you do want to peek under the hood.

Sauna by WordwareElevenLabs

Axial maps how a student thinks - connected cognitive skills across six domains, scored from a diagnostic, fully explainable: every insight traces through typed skill-to-skill dependencies (the "because-trail"). No AI decides a score. Axial Coach connects that model to Sauna as a custom MCP connector, so Sauna can reason and act on a student's real cognitive profile from any surface - voiced with ElevenLabs. What it does - Explains, grounded: "what should this student work on first, and why?" → Sauna answers from the real leverage + because-trail, naming the specific skills a focus unlocks. - One model, two interfaces: - 🎮 Student: a gamified weekly quest: small daily wins, XP, a streak, encouragement anchored to a real strength (confidence through constant small wins). - 🌱 Parent: a holistic report across all six thinking domains + the week's focus and its cross-domain ripple (evidence of broad development). - Runs proactively: a Sauna scheduled task re-pulls the live profile every week and regenerates both - working while you're offline. - Speaks: the coach answers in voice (ElevenLabs). How it's built: The Axial engine runs once, offline; its output for one sample student is cached and served over MCP by a small server. Sauna connects via one URL and discovers six read-only tools. The engine and taxonomy never leave Axial. Sponsors: Sauna (MCP connector, proactive scheduling, two-audience coaching, voice) · ElevenLabs (voice). IP note: Axial's engine, full taxonomy, and cited research are proprietary and not included. This repo contains only the Sauna integration and a citation-free snapshot for one sample demo student - no real data.

Sauna by WordwareElevenLabs

Gmail-overlay Chrome extension (or Sauna App) that rewrites drafts into theatrical personas, restores the original draft, and performs the result with ElevenLabs voice playback.

Sauna by WordwareElevenLabs

katmel

turn any wikipedia article into a personalized podcast for your listening pleasure

Sauna by WordwareElevenLabs

Detsee — what it is and how it works The product (one paragraph) Detsee is a single-user dashboard that calls your overdue customers on the phone using your own cloned voice, captures a payment promise, and atomically writes the promise back to your Google Sheet, drafts a Gmail confirmation, and creates a Calendar follow-up reminder — with you approving every step. You never have to dial yourself; the AI does it, sounds like you, and you review what it captured before anything gets saved. How the pieces connect Your browser ↓ HTTPS Sauna App backend (Detsee) ← hosts the dashboard, the 5 API endpoints, and the audit/snapshot database ↓ authenticated calls (proxy-injected) Five connected services: ├── ElevenLabs — runs the AI agent (your voice, your script) ├── Twilio — places the outbound phone call ├── Google Sheets — source of truth for customer records ├── Gmail — drafts the confirmation email └── Google Calendar — sets the follow-up reminder The five connections all show up in your Sauna Connections panel, pinned to the app at deploy time. The app can't broaden scope at runtime — what you connected is what it uses. How I used ElevenLabs (the AI side) ElevenLabs is where the actual conversation happens. I touched three things on ElevenLabs: The agent — agent_4401kxmgf337fzar0czqtjqj6 ("Payment Recovery Assistant"). Its system prompt was already in place from earlier work: disclosure ("I'm Ali's AI assistant"), a 9-step conversation flow (confirm identity → confirm good time → small talk → ask about balance → ask when they can pay → capture amount + date → read it back → mention payment link → thank and end), hard rules (never pressure, never pretend to be Ali, never send bank info), escalation triggers (flag for owner approval if dispute / payment plan / "I already paid" / "wrong number"). I didn't change any of that. The voice — PATCH'd the agent's TTS config to use your Ali Amjad voice clone (H0yzszDSM9MlKCIvmNrF) instead of the stock default. Now when the agent speaks, it sounds like you, not a stock robot. The phone line — POST /v1/convai/phone-numbers to import your Twilio credentials (AC82d2901fb370bbd3a740a4804831882f + auth token) into ElevenLabs. ElevenLabs now owns the phone_number_id (phnum_1001kxpfzx0bfsmt8mn40p3fmtc4). When Detsee wants to make a call, it asks ElevenLabs: "Twilio, please dial +1 709 765 6030, agent phone_number_id phnum_1001...". ElevenLabs orchestrates Twilio and runs the conversation on the audio side. When a live call ends, ElevenLabs' GET /v1/convai/conversations/{id} returns a structured post-call analysis (11 fields: amount_promised, promise_date, dispute_raised, customer_sentiment, owner_approval_required, etc.) plus the full transcript. Detsee polls this endpoint every 2 seconds while the call is in progress to stream transcript bubbles. How I used Sauna (the platform side) Sauna is the runtime. Detsee is a Sauna App — a long-lived service with its own URL, its own SQLite database, and its own React frontend. The deployment model: Frontend (public/index.html, src/client.tsx) — single React page, three views (dashboard / live call / review), one CSS file. I built the entire UI: cream background, single green accent, hand-drawn SVG icon, 3-column layout (People to Call list · Live Call center · What Happened Automatically right column), inline-SVG mode toggle (Live vs Simulate). Backend (src/handler.ts) — Hono on Cloudflare Workers. Five endpoints: GET /api/customers — reads the Sheet, captures a one-time snapshot for Reset POST /api/call/start — kicks off simulate-conversation or twilio/outbound-call based on mode GET /api/call/{id}/status — polling endpoint, returns transcript + status POST /api/postcall/commit — atomic Sheet write + Gmail draft + Calendar event, refuses if approval needed but not given POST /api/demo/reset — restores Sheet from snapshot, clears audit log Plus GET /llms.txt — the API contract for any agent that lands on the URL Database (src/schema.ts + Drizzle) — two SQLite tables: audit_log (every action with timestamp + outcome) and demo_snapshot (the original Sheet state, captured once, used by Reset). Connections pinning — at deploy time I passed the 5 OAuth accounts to app_deploy. The proxy injects auth headers automatically on every outbound call. The only place I had to manually set X-Sauna-Connection-Id was the Calendar endpoint because Calendar/Gmail/Sheets all share www.googleapis.com and the proxy can't auto-route. Verification — after every meaningful edit I ran app_deploy, browser_verify (snap the URL to confirm it renders), and app_logs (check for runtime errors). The golden loop per the apps skill. The connectors I wired up (in order) # Service What it does Account Connection ID 1 ElevenLabs AI agent + voice + outbound call orchestration ali.amjad52114@gmail.com (via Pipedream OAuth) conn_pd_apn_ygh70LA 2 Google Sheets Customer records (read-only A–H, write I–Q) ali.amjad52114@gmail.com conn_pd_apn_Dph176K 3 Gmail Confirmation email drafts (no send endpoint exists) ali.amjad52114@gmail.com conn_pd_apn_dbhwE1Z 4 Google Calendar Follow-up reminder on the promised date ali.amjad52114@gmail.com conn_pd_apn_r5hX7VD 5 Twilio Outbound phone carrier (verified trial account) ali.amjad52114@gmail.com conn_pd_apn_kVhEXOK What's where on disk apps/payment-recovery-demo/ ├── app.md # manifest — name, description, what the app is ├── package.json # hono + drizzle + react + react-dom ├── public/ │ ├── index.html # shell page + all CSS (cream/green palette) │ ├── detsee-icon.svg # favicon + brand mark (24x24, hand-drawn) │ └── (favicon.png deleted — was broken) ├── src/ │ ├── schema.ts # 2 tables: audit_log, demo_snapshot │ ├── db.ts # sqlite-proxy adapter for Drizzle │ ├── handler.ts # Hono backend — 5 endpoints + /llms.txt │ └── client.tsx # React frontend — 3 views + ErrorBoundary └── migrations/ ├── 0000_init.sql # generated from schema.ts ├── migrations.js └── meta/ Honest boundary Per Claude Code's source-of-truth doc, the ElevenLabs agent prompt, the 11 extraction fields, the 8 dynamic variables, and the Sheet schema were already wired before I touched anything. What I built is the Sauna App: the React dashboard, the Hono backend, the database, the ErrorBoundary, the dual-mode live/simulate switch, the /llms.txt contract, and the integration glue that connects ElevenLabs → Twilio → Sheet → Gmail → Calendar into one click. What Claude Code built is the underlying ElevenLabs configuration and the Sheet column structure. Both halves have to work for Detsee to function. I tested the back half (backend writes confirmed real, Sheet/Gmail/Calendar verified) end-to-end via API. The front half (live human conversation + transcript stream) is wired and Twilio-tested once, but you still need to be the human on the other end of the phone to fully prove it end-to-end.

Sauna by WordwareElevenLabs

The Saunatics!

Using ElevenLabs for text->speech generation (questions) and speech->text transcription (answers), we conduct a survey and use sentiment analysis to see if the user is being truthful, using the Sauna Multimodal LLM. We envision industry applications, while delivering results in a colorful way.

Sauna by WordwareElevenLabs

Compare insured companies' real activity against their current policy via an AI voice interview.

Sauna by WordwareElevenLabs

mock tech interviewer https://mock-interview-gaz2ecdi.sauna.new/

Sauna by WordwareElevenLabs

Calm Team

AI AAC that gives a voice to non-speaking kids a voice

Sauna by WordwareElevenLabs

# Homerun.ai Homerun.ai is a voice-first AI chief of staff for modern families. Built on top of Sauna's persistent memory platform and ElevenLabs' conversational voice technology, Homerun.ai helps families stay organized by remembering everything that matters—school deadlines, family schedules, finances, groceries, healthcare appointments, and relationships. Unlike traditional calendars or task managers, Homerun.ai understands how a household operates over time. It learns responsibilities, routines, preferences, and family dynamics to proactively surface what needs attention before things fall through the cracks. Imagine asking: * "What are we forgetting this week?" * "Who is picking up Maya from soccer on Friday?" * "Do we have enough groceries for dinner tonight?" * "How much did we spend eating out this month?" * "When did we last visit my parents?" * "Find a movie everyone in the family will enjoy during dinner." Homerun.ai connects to school portals, calendars, bills, financial accounts, and household tools through MCP integrations to provide a unified view of family life. Key capabilities include: * Household memory and shared context * School portal notifications and reminders * Calendar synchronization across family members * Monthly expense tracking and bill management * Grocery and meal planning * Parent and caregiver coordination * Friend and relationship management * Voice-enabled interactions powered by ElevenLabs * Personalized family briefings and proactive recommendations The platform organizes information into three relationship layers: 1. Family members living together. 2. Parents and in-laws. 3. Friends and extended relationships. Every morning, Homerun.ai delivers a personalized briefing of upcoming events, deadlines, bills, and important family commitments, ensuring that everyone stays informed and aligned. At its core, Homerun.ai is built around a simple belief: > Running a family is one of the most complex project management problems in the world, and every family deserves an AI that remembers what matters. Homerun.ai transforms household management from a collection of disconnected apps into a single intelligent memory system for the entire family.

Sauna by WordwareElevenLabs

We are home

## Personalized Voice Memory Homerun.ai extends beyond a traditional family assistant by introducing personalized, voice-based family interactions powered by ElevenLabs. Each family member can optionally create a secure voice profile that enables Homerun.ai to deliver reminders, updates, and notifications in the voice of the person who requested them. The goal is not to imitate family members for entertainment, but to preserve context, intent, and emotional relevance in everyday household communication. For example: * If your spouse asks you to call the electrician, Homerun.ai can remind you later using her voice: > "Hey, don't forget to call the electrician before 5 PM today." * If your child records a reminder before leaving for school: > "Dad, remember it's my science fair tomorrow!" * If your parent asks for help scheduling a doctor's appointment: > "Can you please help me book my appointment this week?" By combining persistent memory from Sauna with conversational voice capabilities from ElevenLabs, Homerun.ai creates a more natural and human experience for household coordination. Key capabilities include: * Optional voice profiles for household members. * First-person reminders and notifications. * Context-aware voice briefings. * Voice-based household announcements. * Personalized daily and weekly summaries. * Voice interactions for children, caregivers, and elderly family members. * Multi-user memory with appropriate privacy and permission controls. This feature is particularly valuable for busy families, caregivers, and multigenerational households, where reminders carry more meaning when they come from the people who matter most. Homerun.ai doesn't just remember tasks—it preserves the human context behind them. https://homerun-lyaa7s7s.sauna.new/

Sauna by WordwareElevenLabs

Hacker Band

Voice accessibility companion For ADHD/early-memory-loss users. Sauna remembers everything; ElevenLabs is the warm voice that recalls names, reminders, yesterday's conversations.

Sauna by WordwareElevenLabs

Chinese

Coming straight from G's hot tub. https://sauna-fm-in6hvqaf.sauna.new/

Sauna by WordwareElevenLabs

Icecream

**An AI that fights your consumer-admin battles by phone — cancel subscriptions, dispute charges, and claw back refunds you're owed — and never caves to retention scripts.** You describe an errand in a ChatGPT-style chat ("cancel my gym membership, don't let them upsell me"). The AI asks a couple of clarifying questions, then places a real phone call on your behalf. You can watch the live transcript or ignore it. When it hits a decision beyond its authority, it messages you right in the chat and waits — and if you don't reply in time, it proceeds on your standing instructions. When it's done, you get the full transcript plus a summary. The twist: it **learns your authorization boundaries**. The first time it asks "store credit or hold out for a full refund?", you answer once — and it remembers. Next time, it decides on its own and doesn't interrupt you. Over time it knows exactly what it may do alone versus what needs you. ## How it works - **ElevenLabs** gives the agent a phone — real outbound calls, live voice conversation, real-time transcript. - **Claude (Opus 4.8)** runs the text intake, distills the task, and writes the outcome summary. - **Sauna** gives it a relationship with you — it proactively surfaces the errands worth a call, triggers the assistant at the right time, and stores your authorization boundaries as long-term memory that syncs back after every call. > ElevenLabs gives the agent a phone. Sauna gives it a relationship with you. ## The delegation loop 1. Sauna discovers an open loop (a charge to dispute, a subscription to cancel) and triggers the assistant. 2. The AI checks your authorization memory — if a rule covers the situation, it acts autonomously. 3. It calls the business, negotiates, and only calls *you* back when a decision truly exceeds its authority. 4. The outcome and any newly-learned boundaries flow back into Sauna's memory. ## Stack Next.js + TypeScript · ElevenLabs Conversational AI (Twilio outbound) · Anthropic Claude Opus 4.8 · Sauna (scheduled triggers + long-term memory). Runs end-to-end in a mock mode with zero credentials for reliable demos, and switches to real phone calls when API keys are set. ## Why it matters Every errand that takes five minutes but stays on your to-do list for three weeks — because it requires a phone call — is a battle you're structurally disadvantaged in: you get put on hold, upsold, and worn down. Call Me Maybe compresses all of that into the one or two decisions only you can make.

Sauna by WordwareElevenLabs

Hammer&Chisel

Calorie counting for data-driven and professional athletes who already weight and make their own meals. A conversation in the app captures and logs the caloric and macronutrient information at a high degree of precision that is a requirement for high-performance athletes.

Sauna by WordwareElevenLabsSupabase, Sonnet

s

Bane

this is an accountability partner interacting entirely over the phone. the user gets an introductory phone call to establish what their accountability needs are, including intervals for checking in, defaulting to 20 minutes (pomodoro). Follow up phone calls gently query the user on their progress and offer them a choice of verbal dopamine hits (nag, nice, fun info or story).

Sauna by WordwareElevenLabstwilio

Every piece of data should be secure and private when interacting withwith AI. Tokenization of voice files where the user keeps the merkle root in a private vault, AI agents get their own merkle tree, and then building out a merkle tree chain of custody graph allows for users to maintain privacy and companies like Sauna and ElevenLabs to continue building. https://voicing-fco-o5gx7ixs.sauna.new/ I apologize for the submitted video. You can find a demo video in the github and web app for a sauna and elevenlabs driven voiceover. https://voiceworks-ygitm4zl.sauna.new/demos/voiceworks_demo.mp4

Sauna by WordwareElevenLabs

What it is Most AI apps answer questions. One Honest Question asks the one question you actually need to hear. The experience takes about two minutes: An AI generates one thoughtful self-reflection question. You answer by voice. The AI asks one personalized follow-up based on your response. It distills everything into: One Truth — the insight you've been avoiding. One Action — a concrete step you can take today. ElevenLabs reads the result back to you in a natural voice. The goal isn't to give advice—it's to help people hear themselves more clearly. Why I built it People don't usually need more productivity tips or motivational quotes. They often already know what they should do—they just need the right question. I wanted to explore how voice AI can create a moment of genuine self-reflection in under two minutes. Key features AI-generated reflection questions Voice-first conversation Personalized follow-up questions Concise AI insights Actionable next step Natural voice synthesis with ElevenLabs

Sauna by WordwareElevenLabsreact

canvas

**MyStory** transforms children's drawings into personalized, AI-powered storybooks. After uploading a drawing, the app identifies the characters and creates a unique story, complete with beautiful illustrations, expressive voice narration, and immersive background music. Parents can customize the story's theme, genre, illustration style, and safety settings through a dedicated Parent Mode. The result is a magical storytelling experience where every child's imagination comes to life.

Sauna by WordwareElevenLabsGoogle Gemini Image Model

Give it a description.

Sauna by WordwareElevenLabs