Problem

Modern AI agents, collaborative applications, and other real-time workloads are difficult to run on conventional request-oriented backends because developers must separately assemble persistent state, process lifecycle management, WebSockets, queues, timers, retries, and observability. Rivet targets this infrastructure gap with a stateful runtime in which each agent, session, user, document, or tenant can have a durable process. Its strongest use case is an AI agent that needs persistent context and memory, can remain active across interactions, and can schedule tool calls reliably.

The economic pain is that these workloads are often bursty but need to feel continuously available. Rivet addresses that by letting workloads sleep when idle and scale to zero, while claiming approximately 6 ms cold starts, 22 MB of RAM, and infrastructure costs substantially below full sandbox-based execution. The same model also applies to collaborative documents, chat rooms, per-tenant applications, and multi-step workflows that must survive failures or span long periods.

Product / Service

Rivet provides Actors, a primitive for stateful workloads, together with agentOS, an open-source operating system for agents. Actors colocate in-memory state with compute, support long-lived execution and hibernation, and include real-time WebSockets, workflows, queues, scheduling, and observability. Each AI agent can run as its own actor with persistent context, memory, and scheduled tool calls. agentOS adds an agent runtime that can be embedded in an existing backend, with controlled access to tools and files and a security model based on WebAssembly and V8 isolates.

The delivery model is deliberately flexible. Developers can use RivetKit as a library and run locally in-process, connect to fully managed Rivet Cloud on a global edge network, deploy through a bring-your-own-cloud model, or self-host the Apache 2.0 software in a VPC, on-premises environment, or air-gapped network. This gives Rivet a usage-based cloud business, an enterprise deployment path, and a portability proposition for customers that want to avoid infrastructure lock-in.

Market

Rivet competes in the emerging infrastructure market for AI agents, stateful serverless applications, real-time collaborative software, and durable execution. The clearest named comparison in the available evidence is Cloudflare Durable Objects: Rivet has been described as open-source actor infrastructure similar to that product. It also sits adjacent to AI-agent cloud and sandbox providers such as E2B, although Rivet's differentiation is the combination of durable state, long-lived actors, workflow primitives, and self-hosting rather than only isolated code execution.

Rivet appears commercially launched rather than merely pre-revenue: its offering includes free and paid cloud plans, usage-based compute and storage, and enterprise editions, while the open-source repositories provide visible developer traction. The main Actors repository reports 5,746 stars and agentOS reports 4,265 stars; Rivet also states that it is backed by Y Combinator and a16z Speedrun. A 2024 Show HN discussion claims Rivet already served millions of monthly active users in production, but that is a secondary, self-reported data point, and the available evidence does not provide audited revenue, customer counts, or a confirmed financing total.