Skip to main content
Naïve gives every AI agent its own real-world agent profile: a verified business identity (KYC/EIN/formation), a spend-capped card, an inbox and phone number, and (optionally) a place to run — provisioned per end-user, governed on every action, and instantly revocable. The unit is the bundle: identity + money + comms + runtime as one per-tenant agent profile, governed together — each of your customers’ agents gets its own incorporated entity, inbox, number, card, and runtime as one governed unit.

How it’s organized

Your organization is what you signed up as; it holds the API keys. A project is a scope inside it, holding account kits (policy) and child projects — one per end-user, each with its own agent profile and primitives. Every organization has a default project, so you only meet the layer when you want a second scope. See Projects.

Works with the runtime you already use

Naïve is runtime-agnostic. Provision an agent profile, pull its governed tools, and run the agent inside any agent framework, your own harness, or Naïve’s hosted containers. Frameworks give an agent durability and a sandbox; Naïve gives it a legal identity, a card, and registered comms — and governance (spend caps, approvals, audit, revoke) holds no matter where the agent runs.

CLI First

What’s Different

  • The full per-tenant agent profile bundle — its own legal entity + comms + money + runtime, governed as one. Not just a card attached to someone’s wallet.
  • The governance gateway — every regulated action (a charge, a send, a refund) routes through a policy proxy: spend caps, approvals, audit, and one-click revoke at the tool-call boundary, across the whole agent profile. (Where it binds, and the two places it does not.)
  • Per-tenant comms — phone numbers, A2P/10DLC registration, custom-domain email with isolated sending reputation.
  • Self-service onboarding — agents can register, link accounts, and mint API keys without human intervention.
  • Identity-aware — every primitive knows what resources the agent profile has (“what inboxes do I own?”).
  • MCP-native — hosted MCP server runs alongside REST on the same instance.
  • Agent-discoverableGET /skill.md returns everything an agent needs to get started.

Three Ways to Connect

REST API

Standard HTTP with Authorization: Bearer nv_sk_...

MCP Server

SSE transport at /mcp/sse — works with any MCP client.

CLI

naive command for terminal workflows and agentic use

Orchestration (optional)

Naïve can also host and coordinate agents directly. This is controlled autonomy: humans stay on the exceptions via approvals and audit, not “no humans in the loop.” It’s optional — most teams adopt Naïve for the regulated agent profile bundle first and bring their own runtime.
Two runtimes, and the table below is the older one. The surface in this table — a persistent CEO gateway that decomposes objectives into tasks and dispatches them to AI employees — is the legacy orchestration runtime. It is frozen: every command and route keeps working and accepts no new capabilities.New work declares a team({ lead, agents }) on runtime.durable() and sends it work with naive teams submit. Start at Teams, which also states exactly which of its operations are served on this API build and which answer 501.

Available Primitives

Architecture

What’s Available

Once connected, your agent has access to all available MCP tools and a full REST API. Here are some things to try:

Next Steps

Quickstart

Register and make your first API call in 2 minutes

Orchestration

Set up your AI workforce with CEO, employees, and kanban

Authentication

API key format, auth flow, and error handling

Credits

Sync vs async billing, dynamic model pricing