Rivet Launches Orchestrator to Scale Agentic Workloads via S3

The team behind Rivet, a YC F25 startup, has introduced a new high-density, scalable orchestrator designed specifically for agentic workloads. Built using the actor model, Rivet aims to solve the inefficiencies inherent in running agents within traditional, resource-heavy sandboxes. By utilizing lightweight actors backed by S3, the platform offers a more efficient alternative for developers managing complex AI agents.

Unlike traditional sandboxes that can consume significant memory and suffer from slow startup times, Rivet’s actors are designed for speed and efficiency. The platform provides long-lived processes, in-memory state for context, and durable SQLite storage. According to the founders, this architecture allows for memory usage as low as 72 KB, compared to the hundreds of megabytes typically required by sandboxes. The system is also capable of sub-millisecond durable writes, making it highly responsive for real-time applications.

The choice of S3 as the primary storage medium is a key differentiator for Rivet. While many systems rely on expensive SSDs or high-throughput databases like Postgres, Rivet leverages S3 to keep costs low while maintaining performance. By using an SSD-based caching tier, the system ensures that data is batched to S3 in the background, preventing the storage layer from becoming a bottleneck. This approach allows users to achieve SSD-like speeds at a fraction of the cost of traditional block storage.

Rivet is designed to be flexible, supporting both self-hosted deployments via a single Rust binary and managed cloud options. The platform is already being utilized by companies to orchestrate multiplayer agents and complex workflows. With its open-source model and emphasis on developer experience, Rivet is positioning itself as a robust infrastructure layer for the growing ecosystem of AI agents.

Sources: Hacker News

Aggregated summary. Original: Hacker News