Memories.ai LVMM 2.0 / Project LUCI

https://memories.ai/blog/collaboration-with-qualcomm

Developer-first reference platform for AI wearables. LVMM 2.0 captures visual data through video, encodes it into frames on-device, and stores moments as searchable memories; companion app surfaces past conversations/decisions/experiences. Hardware-level security via Qualcomm partnership; cloud sync for enterprise tiers.

At a glance

Type
On-device Large Visual Memory Model for wearables
Tier
T2
Created
2024 (Memories.ai founded); 2025-07 LVMM 1.0; 2026-01 (CES) Project LUCI + LVMM 2.0
Latest release
not applicable — not OSS
License
not applicable — not OSS (proprietary model)
GitHub
not applicable — no GitHub repo (closed model)
Pricing
searched not found
Funding
$16M total ($8M seed Jul 2025 + $8M extension); Susa Ventures led; Samsung Next + Fusion Fund + Crane + Seedcamp + Creator Ventures

Taxonomy

storage
vector
retrieval
similarity
persistence
long-term
update
append-only
unit
episode
governance
user-controllable
conflict
none

When to use

Optimised for: on-device visual memory encoding for wearables; episodic recall of past conversations/decisions/experiences; privacy via local processing

Anti-fit: not for cloud-only / non-wearable contexts; not for text-only memory (visual focus); not for users requiring open model weights or auditable consolidation

Pros & cons

Pros

On-device visual encoding addresses cloud-bandwidth/privacy concerns of Humane AI Pin / Rabbit R1; Qualcomm + Nvidia Metropolis partnerships; ex-Meta-Ray-Ban founders bring direct wearable AI experience.

Cons

Closed model (no LVMM benchmarks public); pricing/SDK terms undisclosed; reference-platform model relies on third-party hardware partners shipping; LVMM 2.0 Qualcomm rollout ""later this year"" (2026) — not yet shipping at scale.

Claims & capabilities

World's first Large Visual Memory Model (vendor positioning); on-device inference on Qualcomm processors; Nvidia GTC collaboration uses Cosmos-Reason 2 + Metropolis

Technical surface

API surface
Developer SDK (positioned ""like Nexus phones"" for Android-style ecosystem)
Backend storage
Qualcomm SoC + on-device storage + cloud sync
Deployment
Hybrid (on-device LVMM 2.0 + cloud companion)
Embedding model
proprietary LVMM 2.0 visual encoder (no separate text embedder cited)
Multi-tenancy
per-user device
MCP
searched not found
A2A
searched not found
OpenTelemetry
searched not found

Similar systems

Other voice-first / wearable ai memory in the catalog, ranked by inbound references.

  • Bee T1

    Small wearable (pendant / clip / bracelet). Continuous capture; converts speech to text immediately; never stores raw audio . Daily summaries, personal fact sheet, to-do suggestions. Cloud processing on derived text only.

  • Era Computer T5

    Software platform / OS layer for AI wearables (glasses, rings, pendants, speakers) that abstracts hardware from AI orchestration. Memory and model providers are pluggable and user-controlled rather than cloud-locked; supports 130+ LLMs from 14+ providers. Founded 2025 by ex-Humane team.

  • Friend T1

    Pendant necklace (~2 inches). Always-on listening feeds a persistent AI companion persona. Companionship-framed rather than productivity / recall. Cloud LLM backend; no detailed encryption claims.

  • Granola T1

    Mac/iOS app (no hardware). Captures meeting audio via system-audio tap (no bot joining); generates notes from your shorthand + audio. Local audio processing claim; cloud for summaries. Enterprise tier opts out of model training.

  • Limitless T1

    Continuously captures screen + audio + wearable input; AI search across the timeline. Privacy-first (local-first storage). Originated as Rewind macOS app; pivoted to wearable + lifelog product.

  • Omi T1

    Wearable orb (necklace or clip) with open-source hardware + firmware. Continuous audio capture; real-time GPT-4o transcription; searchable memory graph. Supports app screen capture too. Hybrid privacy: cloud by default (SOC 2/HIPAA/AES-256), full local/self-hosted available under MIT.

Row last verified 2026-05-14. Catalog data is CC-BY-4.0 — see how to read this.