Loading prices…
STKR NewsSTKR News0 of 3 free this month
Startups

EQT and FirstMark back trio of H Company alumni in $30m seed round

A group of former H Company researchers just raised $30 million for Remanence, a startup building agentic AI that remembers context and learns from its own history.

Originally on Sifted →
AB

Adrian Boysel

Contributor

Nov 17, 2026

5 min read

Photo illustration / STKR News

The $30 Million Bet on Memory

We are currently living through the gold rush phase of the large language model era. But if you talk to anyone actually building tools on top of these models, you know the dirty secret: they have the memory of a goldfish. You give a model a prompt, it gives you an answer, and for the most part, it starts from scratch the next time you show up. This lack of persistent context is the wall that most AI agents are hitting right now.

A new startup called Remanence just pulled in $30 million in seed funding to fix this specific problem. Led by a trio of alumni from H Company—the French AI lab that made headlines for its own massive $220 million seed round and subsequent internal drama—this new team is pivoting away from general-purpose models toward something far more practical for builders: agentic memory.

EQT Ventures and FirstMark Capital are leading the round. In a market where seed rounds are usually $2 million to $5 million, a $30 million check tells you two things. First, investors are still desperate to find the next foundational layer of the AI stack. Second, they are betting heavily on the specific pedigree of the founders: Karl Tuyls, Daan Wierstra, and Julien Perolat.

Why Context is the Real Bottleneck

For a founder trying to build an automated workflow, the biggest frustration is the "reset button." Most current AI agents operate in a vacuum. They don't remember that you prefer your code in Python, or that your database schema changed last Tuesday, unless you feed that information back into the prompt every single time. This is what we call the context window problem, and it is expensive and inefficient.

Remanence is working on what they describe as agentic AI that can learn from its own history. Instead of just processing tokens, they want to build a system that retains knowledge across sessions. This isn't just about storing a chat log in a database; it is about building a model that updates its internal understanding based on past successes and failures. In plain English: an AI that actually gets better at its job the more you use it.

The H Company Shadow

It is impossible to talk about Remanence without mentioning H Company. For those who didn't follow the saga, H Company was the darling of the European AI scene until a massive rift between the founders and the CEO led to a high-profile split. The three founders of Remanence were part of that original core team. Their exit and subsequent fundraising suggest that the vision they had for H Company wasn't being realized there.

From a builder's perspective, this is a positive sign. It indicates that the talent is moving away from the "build a giant model and hope for the best" strategy and toward solving the specific engineering hurdles that prevent AI from being truly useful in a production environment. When the smartest people in the room leave a $200 million project to start a smaller, more focused one, you should pay attention to what they are focusing on.

What This Means for the AI Stack

We are seeing a shift in the venture capital landscape. The hype around "wrappers"—simple apps that just call the OpenAI API—is dying. Investors are looking for the "middle layer." Remanence fits perfectly into this middle layer. They aren't trying to build a better chatbot; they are trying to build the infrastructure that allows agents to function autonomously over long periods.

If you are building in this space, the takeaway is clear: the next phase of AI isn't about raw power or more parameters. It is about persistence. If an agent can't remember what it did ten minutes ago, it can't be trusted to manage a supply chain, write a complex software suite, or handle customer support without constant human hand-holding.

The Skeptic's Corner

Now, let's be honest. Thirty million dollars is a lot of money for a seed round, even with this pedigree. We have seen plenty of well-funded AI startups burn through cash trying to solve the memory problem with RAG (Retrieval-Augmented Generation) or long-context windows, only to find that the latency and cost make the product unusable for small businesses.

Remanence has to prove that their approach to "agentic memory" is actually a technical breakthrough and not just a more expensive way to search a vector database. The challenge for Tuyls and his team will be translating deep research into a developer-friendly API that doesn't cost a fortune to run. Founders don't care about the academic elegance of a model; they care if it works and if it scales.

The Founder's Perspective

If you are a founder, keep an eye on how Remanence rolls out their beta. We are reaching a point where the base models (GPT-4, Claude 3.5, Gemini) are becoming commodities. The real value is going to be captured by the teams that figure out how to make these models work together in a coherent, long-term way. Remanence is essentially trying to build the "hard drive" and "long-term memory" for the AI brain.

Is this a pivot toward sanity in the AI space? Maybe. It’s certainly more grounded than the usual promises of AGI (Artificial General Intelligence) that we hear from Silicon Valley. Solving the memory problem is a discrete, measurable goal. If they succeed, they won't just be another AI company; they will be the plumbing that everyone else has to use.

The Bottom Line

The Remanence seed round is a signal that the "researcher-led" era of AI is maturing into an "engineer-led" era. The focus is shifting from what a model *knows* to what an agent can *do* over time. For builders, this is the right direction. We don't need more models that can write poetry; we need systems that remember how we work.

  • Focus: Solving the lack of persistence and context in current AI agents.
  • Pedigree: Founded by the former core research team of H Company.
  • Capital: A $30 million seed round suggests high confidence in the technical solution.
  • Utility: The goal is an AI that learns and improves through its own historical data.

Read the original at Sifted →

The Brief

Stay Updated on Cutting-Edge Tech

A six-minute morning dispatch on the markets and the technology shaping them.

Free. No spam. Unsubscribe anytime.

Write for STKR

Become a Contributor

Earn $STKR for published stories on markets, protocols, and culture.

  • Earn $STKR for every published piece
  • Editorial support from the STKR desk
  • Byline visibility across the network
  • First look at the upcoming creator program
Apply to Write

Keep reading

All stories

Comments

24 reader responses