Observed Signal · Mar 16, 2026 · Partnership · Source: techcrunch · Impact: 3/5 · Sentiment: Positive
Memories.ai Unveils Visual Memory Tech for Wearables and Robotics
Memories.ai, a startup spun out of Meta in 2024, is building a "visual memory" infrastructure to enable wearables and robotics to record, index and recall video-based memories. At Nvidia’s GTC conference the company announced a collaboration to use Nvidia’s Cosmos-Reason 2 reasoning vision-language model and Nvidia Metropolis for video search and summarization. Memories.ai launched a large visual memory model (LVMM) in July 2025 and has released a second generation; it also built a custom data-collection wearable called LUCI. The company has raised $16 million to date (an $8M seed in July 2025 plus an $8M extension) led by Susa Ventures with participation from Seedcamp and Fusion Fund, and signed a partnership to run on Qualcomm processors later this year.
The article describes a technical partnership between a startup and major platform (Nvidia) and a chipset partner (Qualcomm) to advance visual-memory infrastructure for wearables and robotics — a meaningful development for edge AI and multimodal indexing but not an industry-shifting policy or major platform product deprecation.
Track Seedcamp Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- Memories.ai announced a collaboration with Nvidia to use Cosmos-Reason 2 and Nvidia Metropolis to develop visual memory technology.
- Memories.ai launched its large visual memory model (LVMM) in July 2025 and has released a second-generation LVMM.
- The company has raised $16 million total: an $8M seed in July 2025 plus an $8M extension led by Susa Ventures, with Seedcamp and Fusion Fund participating.
- Memories.ai built a wearable data-collection device called LUCI to record video for model training and does not plan to sell the hardware.
- Memories.ai signed a partnership to run its technology on Qualcomm processors starting later this year.
Connected Companies & Entities
5 Entities mappedRelated Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
Linker Vision Launches Agentic Video Reasoning Platform with NVIDIA
Linker Vision announced an application-driven platform for agentic video reasoning and expanded deployments across Taiwan, powered by NVIDIA technologies. The rollout builds on a 2026 scale-up of the Kaohsiung lighthouse AI infrastructure and extends autonomous video-reasoning capabilities into Taipei City and Taiwan’s Ministry of Transportation and Communications. The platform uses NVIDIA NemoClaw for secure long-running agents, NVIDIA Metropolis VSS Blueprint for scalable video AI, and NVIDIA Cosmos for Physical AI and world foundation modeling. Linker Vision says the system orchestrates compute, AI agents and services across AI Grids to convert real-time video streams from tens of thousands of cameras into operational intelligence for traffic management, hazard and flooding response, passenger-flow analysis and infrastructure coordination. Paul Shieh, Linker Vision’s founder and CEO, framed the work as building interconnected Physical AI environments for autonomous operations.
MemPalace: Open-source Local AI Memory System
MemPalace is an open-source, local-first AI memory system launched April 5, 2026. Co-founded by actress Milla Jovovich and crypto CEO Ben Sigman, the project reached ~48.5k GitHub stars shortly after release. MemPalace provides cross-session persistent memory for AI assistants using a spatial “palace” hierarchy (Wing/Room/Hall/Drawer), a 4-layer progressive loading strategy that boots with ~50–170 tokens, and a zero-LLM write path that stores data locally (ChromaDB) without calling LLM APIs. It implements a temporal knowledge graph in SQLite to avoid stale facts and exposes 29 MCP tools for integration with Claude Code and other MCP clients. The project introduced an AAAK compression format (claimed 30× compression) and faced independent critiques over benchmark methodology, compression trade-offs, and some unimplemented features.
Memoria: Self‑Evolving Personal AI with Memory
Memoria is a production-ready personal AI MemoryAgent built for the Qwen Cloud Hackathon that implements human-like long-term memory: extraction, prioritisation, decay, consolidation, conflict resolution and reflection. It organises knowledge into three tiers (Session Memory in Redis, Personal Memory in PostgreSQL 16 + pgvector with text-embedding-v3, and a Context Archive for full transcripts). The system uses Qwen models (qwen-plus and qwen-max) for extraction and consolidation, a Python FastAPI backend, Celery workers with Redis broker, and a React frontend. Memoria was deployed on Alibaba Cloud (ECS, ApsaraDB, Redis) and provisioned via Terraform; the author reports a benchmarked 77.6% improvement in decision accuracy across 12 scenarios. Planned next steps include voice input, multi-agent collaboration (MCP), a mobile companion, and fine-tuning Qwen for memory tasks.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.