Restate Raises $20M Series A to Build Durable Infrastructure for AI Agents

Restate has raised $20 million in Series A funding to expand a platform designed to keep modern applications and AI agents running through failures. The round was led by Singular, with participation from Redpoint Ventures and Capital One Ventures, bringing the company’s total funding to $27 million.

The Berlin- and San Francisco-based company says the capital will support continued product development and its expansion in the United States, including a new commercial hub in San Francisco. Restate was founded by members of the team that created Apache Flink, the open-source framework that helped make real-time stream processing a core part of modern data infrastructure.

Why AI agents make software failure more expensive

Traditional backend services are built with failure in mind: servers restart, networks drop connections, APIs become unavailable and processes stop midway through a task. Developers typically address those interruptions with retries, queues, checkpoints and custom recovery logic. That work becomes much harder when an AI agent is expected to operate continuously across models, APIs, tools, other agents and human approvals.

An agent that fails after a long sequence of actions cannot always safely begin again. Restarting may duplicate a payment, repeat an external tool call, lose a human decision or discard expensive model output. As Unite.AI has explained in its guide to how AI agents combine models, tools, memory and control loops, reliability depends on the surrounding system as much as it does on the model itself.

Restate’s answer is durable execution: recording completed work and state so an application can resume from the correct point after an interruption. The company argues that this property must extend beyond an individual workflow to the communication and state shared across a distributed application. Stephan Ewen, Restate’s co-founder and CEO, summarized the ambition in the announcement: “durability has to become a property of the infrastructure itself.”

A runtime built around a distributed log

At the center of Restate’s architecture is a purpose-built distributed log paired with an event-driven orchestrator. Rather than placing durable execution on top of a general-purpose database, the log becomes the foundation for execution, state and communication. Restate says this creates a self-contained system that does not require an external database or search service while remaining lightweight enough to use across an application.

That design is intended to give developers reusable building blocks for durable functions, workflows, services and stateful objects without forcing every workload into a conventional workflow engine. It also reflects the founders’ experience building Apache Flink, which became widely used infrastructure for real-time data systems.

The company recently added a Bring Your Own Cloud option that runs a managed Restate deployment inside a customer’s cloud account and virtual private cloud. For enterprises adopting AI agents in regulated or data-sensitive environments, keeping application data inside existing infrastructure can remove an important barrier to production deployment.

Replit provides an early production test

Restate says it is already used by hundreds of companies, including AI businesses and Fortune 500 organizations. One prominent deployment is Replit Agent, where Restate now handles durable orchestration. The announcement says Replit selected the platform as it redesigned the agent’s architecture and needed a runtime that could preserve long-running work while remaining fast and straightforward for developers.

That deployment matters because coding agents can generate thousands of steps, tool calls and state changes during a single job. If the orchestration layer cannot recover cleanly, the cost is not just downtime; it can mean lost progress, repeated actions and a poor developer experience. A production agent therefore needs infrastructure that can remember what happened, distinguish completed work from unfinished work and resume without replaying side effects.

Funding the layer beneath the agent

Much of the current AI market is focused on models, application interfaces and agent frameworks. Restate’s financing highlights a less visible opportunity: the infrastructure required to keep autonomous software dependable once it leaves a demo environment. As agents take on longer assignments and interact with more external systems, recovery, state consistency and observability become foundational requirements rather than optional engineering features.

Henri Tilloy, Partner at Singular, said the investment thesis rests on Restate bringing execution, state and communication together in one runtime that developers can use across an application. The comparison with Apache Flink is deliberate: the investors are betting that the same team can establish a durable-infrastructure layer for agents and distributed backends as real-time systems become more complex.

The $20 million Series A gives Restate more room to develop that layer and build its U.S. presence. The larger test will be whether durable execution becomes a standard architectural choice for AI applications. If agents are expected to work for hours, coordinate with people and software, and survive inevitable outages, the systems that help them remember and recover may prove as important as the models deciding what to do next.