Rivet
Rivet provides a lightweight actor model runtime and WebAssembly-based sandbox for running AI agents and AI-generated applications at scale. Actors hibernate when idle, waking in milliseconds with durable state intact, while consuming zero memory when dormant. The platform includes agentOS, a V8 isolate-based sandbox that runs agents in-process with roughly 22 MB memory per instance.
- Developer Tools
- Artificial Intelligence
- AI Agents
- Software Only
Funding
Raised to date
$500KRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Founder details are not available yet.
Product
Problem
Traditional infrastructure for running AI agents and AI-generated code relies on containers, VMs, or separate sandbox fleets, which carry heavy memory footprints, slow cold starts, and high per-execution costs. This makes it impractical to run millions of long-lived, stateful agents or to deploy AI-generated applications at scale, especially when workloads need to hibernate and resume on demand.
Solution
Rivet provides a lightweight actor model runtime where each actor lives in a worker process, hibernating when idle and waking in 12.3 ms with durable state intact. Actors get a SQLite database and POSIX filesystem tiered to S3, with idle actors consuming zero memory. The platform includes agentOS, a WebAssembly and V8 isolate-based sandbox that runs agents in-process with ~22 MB memory per instance, and Dynamic Apps for deploying AI-generated applications by ID. Typed SDKs for TypeScript, Node.js, Rust, and React let developers integrate Rivet into their existing backends, and the system supports workflows, cron scheduling, multi-region placement, and OpenTelemetry observability out of the box.
Target Audience
Primary customers are developers building coding agents, agent app builders, company-specific agents on internal data, personal agents, and realtime applications who need scalable, stateful infrastructure that runs where their data already lives.
Features
- Actor-per-workload model with hibernation: idle actors release all resources and wake in 12.3 ms p50 with durable state in SQLite and POSIX filesystem tiered to S3
- agentOS sandbox built on WebAssembly and V8 isolates: 92× faster cold starts (4.8 ms p50), 47× less memory (~22 MB per instance), and 254× cheaper than mainstream sandbox providers
- Dynamic Apps library for AI-generated apps: deploy per-user, per-tenant frontends, REST APIs, SQLite, workflows, and realtime multiplayer in 22 MB while serving, zero while idle
- Typed SDKs for TypeScript, Node.js, Rust, and React with end-to-end type safety and React hooks that subscribe to actor state
- Built-in workflow engine with recorded steps, retries, replay, cron scheduling, and actor-to-actor calls across the cluster
- Single Rust binary control plane with no Kubernetes operator, CRDs, or service mesh; open source under Apache 2.0 and self-hostable