Screen-free voice device that records, transcribes, and summarizes conversations in 120+ languages.
Using speech-to-text summarization with speaker identification, NLP topic clustering for dynamic mind maps, and conversation intelligence with action items.

|
Productivity
|
YC W26

Last Updated:
March 19, 2026

Builds a screen-free, AI-powered voice device that records, transcribes, summarizes, and organizes conversations in real time using model-agnostic LLMs (GPT-5, Claude, Gemini), with end-to-end encryption and HIPAA compliance.
Pocket has publicly announced support for 120+ languages, real-time transcription with speaker identification, dynamic mind mapping, and a contact microphone accessory for capturing both sides of phone calls. They've detailed their model-agnostic architecture (GPT-5, Claude, Gemini) and emphasized SOC 2 Type I & II and HIPAA compliance certifications. Hardware pre-orders and a subscription model for advanced features are live on heypocket.com.
Behind the scenes, Pocket's hiring patterns suggest investment in on-device ML inference and edge computing to improve offline capabilities. GitHub and community activity from co-founder Akshay Narisetti's prior open-source project Omi indicates deep expertise in embedded AI hardware. The model-agnostic approach signals they're building an abstraction layer that could rapidly integrate next-gen models. Strong indicators of expansion into healthcare documentation (HIPAA compliance) and enterprise sales workflows. The lean $500K seed and hardware-first approach suggest a capital-efficient path toward product-market fit before a larger raise.
<p>AI-powered real-time transcription and summarization of meetings and conversations, with automatic speaker identification across 120+ languages.</p>
It's like having a perfect note-taker in every meeting who never misses a word and hands you a clean summary the moment it ends.
Pocket's core use case leverages real-time automatic speech recognition (ASR) models to transcribe conversations as they happen, identifying individual speakers through diarization algorithms. Once transcribed, large language models (GPT-5, Claude, or Gemini, selected dynamically based on task suitability) generate context-aware summaries that distill hour-long meetings into concise briefs. The system automatically extracts action items, assigns them to identified speakers, and organizes everything chronologically. This eliminates the need for manual note-taking, reduces the risk of lost information, and ensures every participant—including those who missed the meeting—has access to accurate, structured records. The device's screen-free, always-ready form factor means recording starts with a single tap, removing friction from the capture process entirely.
It's like TiVo for your meetings—except instead of rewinding bad TV, you're replaying the one brilliant idea someone had at minute 47 that everyone forgot by minute 48.
<p>AI-generated dynamic mind maps that automatically organize spoken ideas, decisions, and themes from conversations into visual knowledge graphs.</p>
It turns your rambling brainstorm into a beautiful, organized map of connected ideas—without you lifting a finger.
Pocket uses natural language processing and topic modeling to analyze transcribed conversations and automatically identify key themes, decisions, sub-topics, and relationships between ideas. These are then rendered as dynamic mind maps that users can explore, edit, and share. The underlying ML pipeline performs entity extraction, semantic clustering, and hierarchical topic decomposition to transform unstructured dialogue into structured visual knowledge. This is particularly valuable for product teams running design sprints, founders iterating on strategy, and creative professionals who think out loud. Rather than manually organizing sticky notes or whiteboard photos after a session, Pocket delivers a ready-made conceptual map that preserves the full context of how ideas evolved and connected during the conversation.
It's like having a brilliant intern who listens to your chaotic whiteboard session and hands you a perfectly organized Pinterest board of your own thoughts.
<p>Captures and analyzes in-person and phone-based sales conversations to extract buyer needs, objections, and next steps using AI-powered conversation intelligence.</p>
It listens to your sales calls and tells you exactly what the buyer wants, what worried them, and what you should do next.
Pocket's contact microphone accessory enables recording of both sides of phone calls, while the primary device captures in-person sales meetings. The AI pipeline then applies conversation intelligence techniques—sentiment analysis, intent classification, objection detection, and buyer signal extraction—to produce structured sales briefs. Each conversation is analyzed for key buyer needs, competitive mentions, pricing sensitivity, decision timelines, and explicit or implicit objections. Action items and follow-up recommendations are automatically generated. For sales teams, this eliminates the CRM data entry bottleneck, ensures institutional knowledge isn't lost when reps leave, and gives managers visibility into field conversations without ride-alongs. The HIPAA-compliant, encrypted architecture also makes it suitable for regulated industries where sales conversations contain sensitive information.
It's like having a seasoned sales coach riding shotgun on every call, except this one actually remembers everything and writes it all down perfectly.
<p>HIPAA-compliant AI transcription and summarization of patient-provider conversations to auto-generate clinical notes and reduce physician documentation burden.</p>
It listens to your doctor's appointment and writes up the clinical notes so your doctor can focus on you instead of a keyboard.
Healthcare providers spend an estimated 1-2 hours per day on documentation. Pocket addresses this by recording patient-provider conversations in a HIPAA-compliant, end-to-end encrypted environment and using clinical NLP models to generate structured clinical notes. The AI identifies medical terminology, maps symptoms to standardized codes, extracts diagnoses, treatment plans, medication changes, and follow-up instructions. The model-agnostic architecture allows Pocket to leverage whichever LLM performs best on medical language tasks. Speaker diarization distinguishes between provider and patient dialogue, ensuring accurate attribution. The screen-free form factor is critical in clinical settings where pulling out a laptop or phone can disrupt the patient relationship. SOC 2 Type II and HIPAA certifications ensure the device meets the stringent security requirements of healthcare organizations.
It's like a medical scribe who never gets tired, never misspells "acetaminophen," and never awkwardly stands in the corner of the exam room.
<p>AI-powered capture and organization of thoughts, tasks, and ideas for neurodiverse users (e.g., ADHD) who benefit from externalizing cognitive load through voice.</p>
It catches all the brilliant ideas your ADHD brain fires off before they vanish, and organizes them into something you can actually act on.
Neurodiverse individuals, particularly those with ADHD, often experience rapid ideation, difficulty with working memory, and challenges translating thoughts into organized written plans. Pocket serves as an always-available cognitive prosthetic—users simply speak their thoughts, and the AI transcribes, categorizes, prioritizes, and structures them into actionable lists, mind maps, and reminders. The screen-free design is intentional: it removes the distraction of a phone or laptop screen, reducing the risk of context-switching that derails focus. The LLM layer understands conversational, non-linear speech patterns and can impose structure on stream-of-consciousness input. This use case represents a significant accessibility and inclusion opportunity, turning Pocket from a productivity tool into a daily-use assistive device that helps users manage executive function challenges with zero friction.
It's like having a personal assistant whose only job is to follow your brain around with a butterfly net, catching every idea before it flutters away.
Akshay Narisetti built and open-sourced Omi, one of the first AI-powered note-taking hardware devices, giving Pocket a massive head start in hardware-software integration, community feedback loops, and real-world edge-AI deployment experience that pure software competitors cannot replicate.