A 19-year-old nabs backing from Google execs for his AI memory startup, Supermemory

Context windows of AI models, which bespeak the quality of a exemplary to “remember” information, person accrued implicit time. However, researchers person suggested new ways to summation semipermanent representation of AI models, arsenic they often can’t clasp discourse implicit respective sessions.

19-year-old laminitis Dhravya Shah is attempting to lick problems successful this country by gathering a representation solution, called Supermemory, for AI apps.

Originally from Mumbai successful India, Shah started gathering consumer-facing bots and apps a fewer years ago. He adjacent sold his bot that formatted tweets into good-looking screenshots to the societal media instrumentality Hypefury.

The founder, who was preparing for an entranceway exam to get into IIT(Indian Institute of Technology), earned bully wealth from this merchantability and decided to determination to the U.S to be Arizona State University instead.

After relocating, helium challenged himself to physique thing caller each week for 40 weeks. During 1 of those weeks, helium built Supermemory (which was initially called Any Context) and enactment it connected GitHub. At that time, the instrumentality allowed you to chat with your Twitter bookmarks.

The existent mentation of the instrumentality extracts “memories” oregon insights from unstructured information and helps the applications recognize the discourse better.

A representation representation extracted by Supermemory Image Credits: Supermemory

Shah secured an internship astatine Cloudflare successful 2024, wherever helium worked connected AI and infrastructure. He aboriginal worked arsenic a developer relations pb astatine the company. During this time, advisors, including Cloudflare CTO Dane Knecht, asked him to crook Supermemory into a product.

This year, helium decided to physique Supermemory full-time.

Supermemory, present described arsenic a cosmopolitan representation API for AI apps, builds a cognition graph based connected the information it processes and personalizes the discourse for the users. For instance, it tin enactment querying crossed month-old entries for a penning oregon a journaling app, oregon searching for an email app. As the solution allows for multimodal inputs, it could besides let a video exertion to fetch applicable assets from a room for a peculiar prompt.

The startup tin ingest immoderate benignant of data, the institution says, including files, docs, chats, projects, emails, PDFs, and app information streams. Its chatbot and notetaker diagnostic lets users adhd memories successful text, adhd files oregon links, and link to apps similar Google Drive, OneDrive, oregon Notion. There’s besides a Chrome hold that lets you easy adhd notes from a website.

“Our halfway spot is to extract insights from immoderate benignant of unstructured information and springiness the apps much discourse astir users. As we enactment crossed multimodal data, our solution is suitable for each kinds of AI apps ranging from email clients to video editors,” Shah said.

Supermemory has secured effect backing of $2.6 cardinal led by Susa Ventures, Browder Capital, and SF1.vc. The circular besides includes idiosyncratic investors similar Cloudflare’s Knecht, Google AI main Jeff Dean, Deepmind merchandise manager Logan Kilpatrick, Sentry laminitis David Cramer, and executives from OpenAI, Meta, and Google.

Shah said astatine 1 point, Y-Combinator besides approached him to articulation 1 of its batches, but helium already had investors connected board, truthful the timing didn’t enactment out.

Image Credits: Supermemory

Joshua Bowder, laminitis and CEO of “Robot lawyer” startup DoNotPay, who runs Bowder Capital arsenic a solo GP, was impressed by Shah’s tenacity.

“I connected with Dhravya implicit X, and what struck maine was however rapidly helium moves and builds things, and that prompted maine to put successful him,” helium said.

The institution has aggregate existing customers, including a16z-backed desktop adjunct Cluely, AI video exertion Montra, AI hunt Scira, Composio’s multi-MCP instrumentality Rube, and Real property startup Rets. Plus, it’s moving with a robotics institution to clasp ocular memories captured by a robot.

While this is simply a slant towards consumers, the app feels much similar a playground for developers to recognize much astir the instrumentality and perchance usage it successful their workflows oregon their ain apps.

Supermemory has important competitors successful the representation space. Startups similar Felicis Ventures-backed Letta and Mem0 (where Shah worked for a abbreviated while) are gathering a representation furniture for agents. Supermemory’s ain backer, Susa Ventures, has invested successful Memories.ai on with Samsung, which tin pat into thousands of hours of footage to get insights. Shah says that portion these startups mightiness service antithetic industries and usage cases, Supermemory volition basal retired due to the fact that it offers little latency.

“More and much AI companies volition request a representation layer. Supermemory’s solution provides precocious show portion allowing you to aboveground applicable discourse quickly.” Bowder said.