Naïve

Naïve is an infrastructure platform for building, deploying and operating AI agents and agent-built applications. Its stack provisions per-agent cloud services — Postgres, authentication, object storage, realtime sync, edge functions, queues and cron — through a single command ("naive up"), alongside hosting for fullstack apps, serverless functions and GPU workloads on managed compute and sandboxed containers.

Beyond the application backend, the platform gives agents a real-world operating presence: KYC/KYB verification, US LLC and EIN incorporation, registered agent services, email inboxes, domains, DNS and phone/SMS numbers, plus distribution to social channels and search. Money movement runs through scoped virtual cards with hard spend limits enforced at card authorization, together with invoicing so operators can bill their own customers. A governance control plane sits in front of every primitive, offering audit logging and replay, capability allow/deny lists per agent and per end-user, session-scoped MCP tool access, human-in-the-loop approval policies for sensitive actions, and mid-flight revocation of cards, sessions and capabilities.

Agents and multi-agent teams are declared as infrastructure-as-code via the @usenaive-sdk/iac package, where identity, permissions, budgets and coordination edges are versioned in the repository. Teams run on Vetta, a durable runtime supporting a lead agent with a typed roster and state that survives crashes, or on Hermes. A model router exposes 300+ models through one API with task-based routing, automatic fallbacks and exact-cost metering, and the platform connects to 100+ third-party tools including Stripe, Supabase, Vercel, QuickBooks, PostHog, Mercury, Rippling, Brave Search and Instantly.

Business model

Naïve sells developer infrastructure delivered as a hosted platform and SDK/CLI, bundling compute, database, identity, payments and model routing primitives behind a single API key and configuration file.