Memories
San Francisco, US · 6 known investors
Memories.ai builds a Large Visual Memory Model and API infrastructure that lets AI systems index, recall and act on video.
Also known as Memories AI · Memories.ai
Founders & leadership
Investors · 6
Also in the syndicate · 1
Company profile
researched Aug 2026Memories.ai is an AI research lab and infrastructure company developing what it calls a Large Visual Memory Model (LVMM), a system that gives AI persistent, long-horizon visual memory. Rather than loading whole clips into a model's context, the LVMM compresses video into structured, queryable representations, indexes them across large volumes of footage, and retrieves relevant moments across long time spans. The company frames its product around three functions: visual understanding and indexing (processing frames, faces, scenes and speech into structured data), persistent visual memory (a searchable knowledge layer of people, events, actions and context that accumulates over time), and agents that act on that memory to perform end-to-end tasks such as flagging safety incidents, auto-tagging libraries and surfacing specific moments.
The technology is delivered as APIs — a Video Intelligence API covering captioning, streaming understanding and human re-identification, and a Visual Search API for hosting, indexing and searching video at scale — alongside a chatbot-style web application where users upload videos or connect existing libraries. Application areas include security and safety (searching months of surveillance footage), media and entertainment (locating scenes or visual elements across large archives), marketing analytics (sentiment and mention analysis across social video), robotics and physical AI, and consumer devices. Deployment options span edge, on-premises and cloud.
The company also runs Project LUCI, a program built around a wearable recording device used by its own data collectors to capture first-person video for model training; management has stated it does not intend to become a hardware company or sell the devices. LUCI products referenced include LUCI Desktop and LUCI Pin, and the company has said LUCI will be presented at Microsoft Build 2026.
Founding story
Co-founders Shawn Shen and Ben Zhou (also referred to as Enmin Zhou) met as researchers at Meta Reality Labs, where they worked on the AI system behind Meta's Ray-Ban smart glasses. That work led them to question how useful AI wearables would be if users could not recall the video data being recorded; after finding no existing visual memory solution for AI, they spun out of Meta to build one. Shen holds a PhD in Computer Science from Cambridge and serves as CEO; Zhou, an engineer with an applied mathematics background, is CTO. The company describes its origins as beginning at Cambridge around the thesis that visual memory deserves its own field of study.
Business model
Memories.ai sells video intelligence and visual memory infrastructure to other companies and platforms, delivered through APIs and a web application, with an emphasis on models and infrastructure rather than end-user hardware. Its engineers work alongside customer teams to bring deployments into production, with edge, on-premises or cloud options that are modular around the customer's stack. Partnerships with platform and silicon providers — including Samsung, Nvidia and Qualcomm — are a channel for reaching device and platform scale.
The platform is offered commercially via API access and a web application, with published pricing and a "Contact Sales" enterprise motion; the company states its commercialization focus is on the model and infrastructure layer.
Traction
By July 2025 the company had indexed 1 million hours of video and listed Samsung, Aosu, PixVerse and Viggle among the companies using its technology. Its site cites more than 100 million cameras reached through partner integrations and more than six global technology partnerships. Total disclosed funding is $16 million. The company reports working with large wearable makers without naming them, and has announced collaborations with Samsung, Nvidia and Qualcomm.
Latest developments
In March 2026 Memories.ai announced a collaboration with Nvidia at the GTC conference, using Cosmos-Reason 2 and Nvidia Metropolis to advance its visual memory stack, and disclosed a partnership with Qualcomm to run its models on Qualcomm processors later in the year. It also released the second generation of its Large Visual Memory Model. The company has announced that Project LUCI will be featured at Microsoft Build 2026, framed around unifying digital and physical memory with on-device AI.
▸Full profile — market position, technology, go-to-market, geography, history, risks & controversies
Market position
The company positions itself as building a missing layer of the AI stack — persistent visual memory — and describes its LVMM as the first of its kind. Its seed investor characterized the approach as "Google Search, but for video," contrasting it with systems that must load entire clips into memory. Existing memory features from OpenAI, xAI and Google have focused mainly on text; Memories.ai targets visual and physical-world applications. Its founder compared the LVMM to a smaller version of Google's Gemini Embedding 2 multimodal indexing and retrieval model.
Differentiation rests on a memory-first architecture that compresses and indexes video for retrieval across years of footage, rather than processing clips in isolation — addressing the limitation that most AI systems lose track of video content after roughly 15 to 60 minutes. The company claims 100x more video memory than previously possible, an owned training-data pipeline via the LUCI wearable, a research-led team including former Meta, Google and Nvidia scientists, and deployment flexibility across edge, on-premises and cloud.
Technology
The core technology is the Large Visual Memory Model, a memory-first retrieval architecture that compresses video into structured representations, indexes them across millions of video hours, and retrieves results across long time horizons instead of loading full clips into context. Capabilities include captioning, streaming understanding, human re-identification, clip search and agentic analysis. A second-generation LVMM has been released, and the company works with Nvidia's Cosmos-Reason 2 reasoning vision language model and Nvidia Metropolis reference architecture for video search and summarization, with Qualcomm processor support planned. Training data is gathered via the LUCI wearable recorder used by in-house data collectors. Published research includes work on structured memory for long agentic video reasoning (MAGIC-Video), metric-aligned long-horizon video memory (SpatialMem), memory-augmented compression and token compression for efficient video understanding (O-MARC, MARC), any-to-any audio-video-text retrieval (OmniRetriever), UGC video captioning, and security video understanding (Memories-S0). The platform is SOC 2 Type II certified.
Go-to-market
The company combines a self-serve developer path (free sign-up, API documentation, published pricing) with direct enterprise sales and hands-on deployment support. It amplifies reach through technology partnerships with device and chip makers such as Samsung, Nvidia and Qualcomm, and builds credibility through published research and conference presence at events including Nvidia GTC and Microsoft Build.
Enterprises and platforms with large video volumes, spanning physical security and safety, media and entertainment production, marketing and social-video analytics, robotics and physical AI, consumer electronics and wearables, telecommunications, and enterprise AI. Named users and partners include Samsung, Aosu, PixVerse and Viggle.
Geography
Headquartered in San Francisco, California, with research roots at Cambridge in the UK and partner relationships spanning global consumer electronics, security, telecommunications and enterprise AI companies.
History
Memories.ai was founded in 2024 by former Meta Reality Labs researchers and is headquartered in San Francisco. In July 2025 it emerged publicly with an $8 million seed round led by Susa Ventures and launched its first Large Visual Memory Model, having by then indexed roughly 1 million hours of video. An $8 million extension brought total funding to $16 million. In March 2026 the company announced a collaboration with Nvidia at the GTC conference, released a second-generation LVMM, and disclosed a partnership with Qualcomm to run its models on Qualcomm processors starting later that year.
Risks & controversies
The founder has said the wearables and robotics markets that the company ultimately targets are not yet mature, so near-term demand is limited and commercialization is concentrated on models and infrastructure. The company also declines to name the large wearable companies it works with, limiting verification of those relationships. Its data-collection approach relies on first-person video recorded by human data collectors wearing the LUCI device.
Compiled by commissioned research from 8 cited public sources — announcements, filings, and press listed under research sources below.
Key figures
latest reportedCompany-reported or press-reported figures, each dated to when it was claimed — not independently audited.
Timeline · 7
launches, deals, and filingsMemories.ai announced a collaboration with Nvidia at the GTC conference, using Nvidia's Cosmos-Reason 2 reasoning vision language model and Nvidia Metropolis reference architecture for video search and summarization to develop its visual memory technology.
The company signed a partnership with Qualcomm to run its visual memory models on Qualcomm processors starting later in 2026.
Memories.ai released the second generation of its LVMM, which its CEO compared to a smaller version of Google's Gemini Embedding 2 multimodal indexing and retrieval model.
The company announced Project LUCI at Microsoft Build 2026, positioned around unifying digital and physical memory with on-device AI. LUCI is a wearable recording device used by the company's data collectors to gather training video, with LUCI Desktop and LUCI Pin products referenced.
Memories.ai announced $8 million in seed funding led by Susa Ventures with participation from Samsung Next, Crane Venture Partners, Fusion Fund, Seedcamp and Creator Ventures. Wilson Sonsini Goodrich & Rosati advised the company on the transaction.
$8M source ↗
Through a partnership with Samsung, Memories.ai is bringing visual memory capabilities to mobile and consumer device experiences.
The company launched its first Large Visual Memory Model, described as the world's first, providing persistent searchable memory for video at scale via API and a chatbot web app.
Dated company events from announcements, filings, and press; legal rows summarize public dockets and regulator releases.
▸Research sources · 8
primary sources listed
- Memoriesmemories.ai · web
8 public sources were cited for this profile; the first-party ones are listed here.
Frequently asked questions
- What does Memories do?
- Memories.ai builds a Large Visual Memory Model and API infrastructure that lets AI systems index, recall and act on video.
- Who founded Memories?
- Memories was founded by Shawn Shen.
- Who are Memories's investors?
- Memories's investors include Crane Venture Partners, Creator Ventures, Fusion, Samsung Next, Seedcamp.
- Where is Memories headquartered?
- Memories is headquartered in San Francisco, US.