Funding details
| Item | Value |
|---|---|
| Company | Remanence (France, Paris) |
| Round | Seed |
| Amount | $30M |
| Lead | EQT Ventures, FirstMark |
| Participants | Zero Prime, Emblem, Drysdale, Global Founders Capital, Yellow; angels include Datadog CEO Olivier Pomel, Mirakl co-founder Philippe Corrot, Pigment co-founder Romain Niccoli, and Synthesia co-founders Victor Riparbelli and Steffen Tjerrild |
| Valuation | Undisclosed |
| Total raised | $30M (first round) |
| Founded | 2026 (this is the company's first public funding round) |
| Headcount | Undisclosed (three co-founders so far; plans to hire primarily in Paris, with early presence in London and San Francisco) |
What the company does
Remanence turns the execution data an enterprise's own AI agents generate into models the enterprise owns — its argument is that the more agents a company deploys, the more execution-trace data it produces, but most companies lack the infrastructure to turn that data into a model they can keep improving, so the data goes to waste and the company stays dependent on external model providers.
The product converts the execution traces from a company's existing AI models into custom training environments, then post-trains open-weight models against them using reinforcement learning — agents practice tasks and make decisions inside a simulated version of the enterprise's own environment, learning from mistakes, with the system adjusting model parameters based on outcomes. As agents keep generating more data, the models can be retrained on a rolling basis, continuously improving accuracy and adapting to changing workflows. Companies can keep the data on their own infrastructure and decide for themselves where training and deployment happen, ultimately owning the resulting model weights and the intelligence derived from them.
The three co-founders — Pierre-Louis Cedoz, Hamza Benchekroun and Axel Moyal — all previously worked together at the French AI lab H Company, where they led post-training on the open-weight Holo models, which the company says outperformed several leading open- and closed-source models on computer-use benchmarks; the team's combined background also includes Google DeepMind, Meta and InstaDeep. Remanence already has design partners trialing the product across legal, customer support and enterprise software in Europe and the US.
What this funding signals
What it means for the agent ecosystem
This is a sign that "enterprise agent data sovereignty" is starting to get priced seriously by capital. As companies deploy more agents, the scale of execution data grows with it, and Remanence is betting that enterprises will eventually prefer turning that data into a model they own and can keep iterating on, rather than continuing to rent capability through generic model APIs — shifting the battleground from "which model is strongest" to "who can turn a company's own data into an asset for it."
What investors are betting on
EQT Ventures partner Tom Mendoza frames the next phase of enterprise AI as being about who owns the intelligence: as companies deploy more agents, they generate more data, but most lack the infrastructure to turn it into continuously improving models, and he's betting on the team's technical depth to close that gap. FirstMark general partner Matt Turck points out that post-training is a skill that today lives almost entirely inside a handful of labs — the Remanence team built Holo at H Company, beating frontier-lab models on benchmarks, and is now turning that capability into something every enterprise can use, which he sees as a scarcer moat than "yet another agent framework."
Numbers worth watching
Flow summarized from Remanence's own statements and Tech Funding News reporting.
- At $30M, this is the smallest seed round among the three companies in this batch, but the angel list — founders of Datadog, Mirakl, Pigment and Synthesia — signals real industry consensus on the problem, not just a financial bet.
- Compared to adjacent plays: Bespoke Labs raised $40M across seed and Series A for "simulated enterprise environments," Mercor raised $750M at a $20B valuation for "agent evaluation and training environments," and River AI raised $1.1B for "fine-tuning and reinforcement learning on open-weight models." Remanence's angle — using a company's own real agent traces as the training environment, rather than a generic simulation — prices differently from its neighbors, making direct valuation-multiple comparisons hard right now.
- The company hasn't disclosed customer counts or quantified trial results; "already trialing with design partners in legal, customer support and enterprise software" is the only verifiable signal of traction so far.
Watchlist status
Remanence is not yet on the watchlist. Recommend adding it under section B4 (Agent Memory / Context) or a new "enterprise agent data sovereignty / continual learning" subcategory, tracking: turning enterprise agent execution data into enterprise-owned post-trained models, $30M seed led by EQT Ventures and FirstMark, in an adjacent-but-distinct lane from Bespoke Labs, Mercor and River AI.
Today's takeaway
Most "agent training infrastructure" companies sell generic simulated environments; Remanence sells "a training environment built from your own agent's real execution data." That suggests the next wave of differentiation may not be whose simulation most resembles the real world, but who can fastest turn a customer's own real execution data into a specialized model the customer can take with them, understand, and keep iterating on.