LoginSign Up
← Back to all sparks
R

Rivet

DEVOPS
Velocity8.8

Rivet positions its Actors runtime as the infrastructure layer for enterprise-ready, AI-native application deployment.

actor-modelbyocmcpagent-infrastructuredynamic-appsserverless
◆Current state
Rivet has shipped three substantive capability moves in rapid succession: BYOC (Bring Your Own Cloud, letting enterprises run Rivet's control plane inside their own AWS or GCP VPCs), MCP integration (exposing Rivet Actors as a first-class tool in Claude Code, Cursor, Codex, and Gemini CLI), and Dynamic Apps (a V8-isolate-based runtime for deploying AI-generated applications for end users). Underneath all of this is the Actors model — a durable, stateful compute primitive built on open-source infrastructure. Durable Streams, a zero-disk SQLite storage engine with S3 tiering, and the agentOS execution API round out the technical foundation.
◆Where it's heading
Rivet is building toward a single answer to a specific question: where does agent-generated, user-facing software actually run? The BYOC move unlocks regulated industries and large enterprises who can't send data to a SaaS control plane. MCP turns Rivet's Actors into something any AI client can discover and call without bespoke integration. Dynamic Apps makes Rivet the runtime, not just the infrastructure, for user-generated software. The through-line is that Rivet wants every AI agent — whether built by a developer or generated at runtime — to run on the Actors primitive with Rivet managing the lifecycle.
◆Prediction
BYOC on AWS/GCP is the foundation; Azure support and SOC 2 certification are the logical next steps to close enterprise deals. Expect MCP to expand to more clients (OpenAI Codex, Copilot, Windsurf) as the MCP ecosystem grows, and Dynamic Apps to get versioning and rollback — the missing piece for user-facing production deployments.

◆Recent moves

  1. 1d ago

    Introducing Rivet BYOC

    ⚡ SPARK

    BYOC changes who Rivet can sell to. Running the control plane inside an enterprise's own VPC eliminates the primary objection from regulated-industry buyers — Rivet's control plane touches no customer data. This is a go-to-market unlock, not just an architectural addition.

  2. 6d ago

    Introducing Rivet MCP

    ⚡ SPARK

    MCP integration makes Rivet Actors discoverable and callable from any AI client that speaks the Model Context Protocol — Claude Code, Cursor, Codex, Gemini CLI, VS Code. This converts Rivet from infrastructure developers configure to infrastructure AI agents can use autonomously, which is the actual surface area being competed over in 2026.

  3. 13d ago

    Durable Streams now supports Rivet Actors

    Durable Streams — Rivet's real-time streaming layer — now runs on Rivet Actors as the underlying primitive, making it self-hostable and open-source. This deepens the Actors abstraction, showing that Rivet's own internal features are built on the same primitive it offers developers.

  4. 16d ago

    Introducing Dynamic Apps: Deploy AI-Generated Apps for Your Users

    ⚡ SPARK

    Dynamic Apps extends Rivet's scope from developer infrastructure to end-user runtime. If Rivet previously built the pipes for agents, Dynamic Apps makes Rivet the place where agent-generated applications actually run — a V8 isolate layer that handles scale-to-zero and serverless lifecycle for software your users, not your engineers, define.

  5. 1mo ago

    Rivet ships zero-disk SQLite with S3-tiered cold storage

    Rivet's storage engine for SQLite now tiers cold data to S3 with zero local disk, keeping writes fast while offloading storage costs. This is a meaningful infrastructure improvement for Actors that need durable embedded databases — the common case for stateful agent workloads — without paying for always-warm block storage.

  6. 1mo ago

    Introducing agentOS Execution API for JavaScript and Python

    agentOS's execution API now supports JavaScript (Node.js) and Python directly, replacing the bash-only interface. For the growing number of agents that need to run language-native scripts rather than shell commands, this is a real usability step — and it extends Rivet's runtime compatibility to the two dominant AI-agent scripting environments.