Mem0 raises $24M from YC, Peak XV and Basis Set to build the memory layer for AI apps

Taranjeet Singh (pictured right) has launched six companies, with immoderate failing and others seeing varying degrees of success. His seventh, Mem0, could beryllium his defining one.

The startup starts with the premise that ample connection models can’t retrieve past interactions the mode humans do. If 2 radical are chatting and the transportation drops, they tin resume the conversation. AI models, by contrast, hide everything and commencement from scratch.

Mem0 fixes that. Singh calls it a “memory passport,” wherever your AI representation travels with you crossed apps and agents, conscionable similar email oregon logins bash today. The YC-backed startup, launched successful January 2024, has raised $24 cardinal ($3.9 cardinal successful antecedently unannounced effect backing and a $20 cardinal Series A.)

AI-focused early-stage money Basis Set Ventures led the Series A, with information from existing investors Kindred Ventures and Y Combinator, arsenic good arsenic caller backers including Peak XV Partners and the GitHub Fund.

Notable angels see Dharmesh Shah (HubSpot), Scott Belsky (ex-CPO Adobe), Olivier Pomel (Datadog), Thomas Dohmke (ex-CEO GitHub), Paul Copplestone (Supabase), James Hawkins (PostHog), Lukas Biewald (Weights & Biases), Brian Balfour (Reforge), Philip Rathle (Neo4j), and Jennifer Taylor (former President, Plaid).

Having respective leaders who helped signifier the modern bundle ecosystem stake connected Mem0 (pronounced “mem zero”) underscores its promise, and the traction from the four-person squad backs it up.

So far, the open-source API, which claims to beryllium the astir wide adopted representation model for AI developers, has surpassed 41,000 GitHub stars and recorded implicit 13 cardinal Python bundle downloads. In Q1 2025, Mem0 processed 35 cardinal API calls. By Q3, that fig jumped to 186 million, increasing astir 30% period implicit month. 

Beyond open-source adoption, much than 80,000 developers person signed up for its unreality service. Mem0’s unreality API present handles much representation operations than immoderate different supplier and serves arsenic the exclusive representation supplier for AWS’s caller Agent SDK.

In aboriginal 2023, Singh was inactive successful Bangalore, India. He started his vocation arsenic a bundle technologist astatine Paytm, 1 of India’s astir invaluable startups, earlier becoming Khatabook’s archetypal maturation engineer. He discontinue successful precocious 2022, conscionable arsenic the ChatGPT question was astir to crest, and built 1 of the archetypal GPT app stores, which scaled to implicit a cardinal users.

That acquisition led him to make Embedchain, an open-source task that lets developers index, retrieve, and sync unstructured data. As the task took off, earning much than 8,000 GitHub stars, Singh sent implicit 200 acold emails to founders, investors, and engineers successful Silicon Valley.

“I reached retired to astir each celebrated tech entrepreneur that you mightiness person heard of and was rather persistent. Some of them responded, and aft proceeding america out, scheduled america to alert from Bangalore to San Francisco wrong 36 hours,” Singh said.  

Once successful the U.S., Singh reconnected with his longtime person and present co-founder and CTO, Deshraj Yadav, who had led the AI Platform astatine Tesla Autopilot. Together, they had antecedently built EvalAI, an open-source Kaggle alternate that grew to 1.6K GitHub stars.

While experimenting with Embedchain, the duo launched a meditation app inspired by Indian yogi Sadhguru. The app went viral successful India, but Singh says users kept sharing the aforesaid feedback: “Hey, I’m connected this meditative journey, but the app doesn’t retrieve that.” So they pivoted from Embedchain to Mem0 to lick that problem.

The thought of representation for AI isn’t new, but it’s rapidly becoming a captious battleground. OpenAI, for instance, began investigating semipermanent representation features successful ChatGPT successful aboriginal 2024, and its CEO, Sam Altman, has hinted that persistent representation volition beryllium cardinal to OpenAI’s upcoming hardware device. Other AI labs are besides launching experimental representation systems for their agents.

Singh argues that portion large AI labs are gathering representation systems, they person small inducement to marque them portable oregon interoperable. “Memory is becoming 1 of their cardinal moats present that LLMs are getting commoditized,” helium said.

He explains that portion consumers tin bask persistent, personalized experiences successful ChatGPT, developers who privation to physique applications — say, a concern companion that remembers a user’s trading past — request an open, neutral solution similar Mem0.

“We privation developers to connection day-one personalization done a shared representation network,” Singh said. “Think of it arsenic Plaid for memory. That’s enactment two. For now, we’re laser-focused connected gathering the champion representation merchandise possible.”

Mem0’s model lets developers store, retrieve, and germinate idiosyncratic representation crossed models, applications, and platforms. It’s model-agnostic, compatible with OpenAI, Anthropic, oregon immoderate open-source LLM, and integrates straight with frameworks similar LangChain and LlamaIndex.

Developers usage Mem0 to make applications that turn smarter with each interaction: therapy bots that callback past conversations, productivity agents that retrieve idiosyncratic habits, and AI companions that accommodate implicit time. Customers scope from indie developers to endeavor teams gathering copilots and automation tools.

“We backed Mem0 from its earliest days — adjacent earlier YC — due to the fact that representation is foundational to the aboriginal of AI,” said Lan Xuezhao, laminitis and spouse astatine Basis Set Ventures. “We’re doubling down arsenic the squad continues to tackle 1 of the hardest and astir important infrastructure challenges: enabling AI systems to physique lasting, contextual memory.”

Other early-stage startups successful the representation abstraction see Supermemory (whose laminitis concisely worked astatine Mem0), Felicis-backed Letta, and Memories.ai.