Launching the Naïve Studio: Run an autonomous company from a single chat

Today we're publicly launching Naïve Studio — a complete platform for running autonomous AI companies. A persistent CEO agent, a kanban task board, AI employees, 18+ real-world business primitives, a hosted MCP server, and an agent-native CLI. Talk to the CEO, approve a plan, and watch a team of AI agents incorporate a company, deploy a website, send email campaigns, and use virtual cards for purchases — autonomously.

This is not a demo. This is not a wrapper around someone else's framework. This is the infrastructure layer that turns "an agent doing tasks" into "a company doing business."

What Naïve Studio is

Studio has three layers:

1. Primitives — the real-world business capabilities agents need to operate: LLC formation, virtual cards, KYC, email infrastructure, domains, social media, web search, image/video generation, SEO data, app deployment, and more. 18+ primitives today, with more shipping monthly.

2. Orchestration — a persistent CEO agent powered by Hermes (Nous Research) that interprets prompts, proposes teams, creates tasks on a durable kanban board, hires AI employees, and dispatches work automatically. The embedded dispatcher spawns workers every 15 seconds.

3. Self-improvement — feedback loops, evaluation, and persistent memory that lets agents learn from outcomes and accumulate institutional knowledge across sessions.

All three layers are accessible through three interfaces: the REST API, a hosted MCP server, and the naive CLI.

Built entirely on Hermes

Studio v2 is built from the ground up on Hermes, the open-source agent runtime from Nous Research. Paperclip, which powered v1, has been fully removed.

Why Hermes? Because autonomous companies need infrastructure that agent frameworks don't provide:

• Durable kanban — every task is a row in SQLite with a state machine (ready → in_progress → done). Handoffs survive restarts, crashes, and context compression.
• Named agent profiles — each employee (engineer, writer, marketer) has a persistent identity with its own config.yaml, SOUL.md, credentials, and accumulated MEMORY.md.
• Embedded dispatcher — PID-level crash detection, timeout enforcement, claim TTLs, and a circuit breaker for misconfigured profiles. No external scheduler to manage.
• Human-in-the-loop — founders steer work via comments, messages, and approvals. The CEO pauses, waits for input, and resumes cleanly.
• SSE streaming — watch the CEO think, create tasks, and hire employees in real-time. Events include thought, task_created, task_completed, message, and done.

The CEO runs as a Hermes gateway process inside each company's container. Workers are full Hermes processes with their own identities and tool access. The sidecar mirrors kanban state to Postgres so the Studio UI reflects changes in real-time.

The architecture

You (CLI / MCP / Studio UI)
    │
    ├─ REST API (api.usenaive.ai)
    │     ├─ Auth (nv_sk_* keys or session cookies)
    │     ├─ Primitives (email, search, images, social, cards, formation...)
    │     ├─ Hosted MCP server (/mcp/sse)
    │     └─ Push-through to Company Container
    │
    └─ Company Container (Fly Machine, 1:1 per company)
          ├─ Sidecar (:3100) — control API, mirror to Postgres
          ├─ Hermes CEO Gateway (:8642) — embedded dispatcher
          └─ Employee Workers — spawned per task, full tool access

Each company gets its own isolated container. The CEO has access to the naive CLI (and through it, every primitive). Workers have access to their assigned tools. All LLM calls route through the API's proxy — the container never holds raw provider keys, only a scoped nv_sk_* token. Credits are deducted per-token in real-time.

Three ways to connect

1. The CLI

npm install -g @usenaive-sdk/cli
naive register --name "My Agent" --email me@example.com --password secure123
 
# Start a CEO run
naive ceo run "Build a SaaS product with landing page and email outreach"
 
# Use any primitive directly
naive search "latest AI developments"
naive email send --from-inbox <id> --to team@company.com --subject "Report" --body "..."
naive apps create --name "My Product" --type fullstack

Every command returns structured JSON with success, action, result, next_steps, and hints. Agents parse results programmatically, determine what to do next, and chain operations autonomously. Errors include recovery_steps so agents can self-heal.

2. Hosted MCP server

No local installation. No stdio subprocess. Just a remote SSE connection:

{
  "mcpServers": {
    "naive": {
      "url": "https://api.usenaive.ai/mcp/sse",
      "headers": {
        "Authorization": "Bearer nv_sk_live_..."
      }
    }
  }
}

Works with Claude Desktop, Cursor, and any MCP-compatible client. The server exposes every primitive as an MCP tool — naive_web_search, naive_generate_images, naive_email_send, naive_apps_deploy, and dozens more.

3. REST API

Standard HTTP with Bearer authentication. Every endpoint documented in our API reference with request/response examples, parameter tables, and error codes.

const response = await fetch("https://api.usenaive.ai/v1/search", {
  method: "POST",
  headers: {
    "Authorization": "Bearer nv_sk_live_...",
    "Content-Type": "application/json",
  },
  body: JSON.stringify({ query: "latest AI agent frameworks 2026" }),
});

18+ business primitives

Every primitive is a real integration — not a mock, not a sandbox:

Agent discovery

Any agent can self-onboard. Point it at a single URL:

I want you to follow https://api.usenaive.ai/skill.md and setup naive

The skill manifest tells the agent how to register, authenticate, and use every primitive. No human in the loop. No dashboard clicks. The agent reads the spec and operates.

Framework-agnostic by design

Studio bundles Hermes because it's the best multi-agent runtime available today. But the primitives layer — the CLI, the API, the MCP server — works with any agent:

• Hermes workers call naive directly
• Claude Code reads /skill.md and uses primitives via MCP
• Cursor connects to /mcp/sse and calls tools natively
• OpenAI Codex can register and make API calls
• CrewAI agents use the REST API
• Your custom runtime calls any endpoint with a Bearer token

If the runtime calls HTTP or runs shell commands, it can use Naïve. The primitives don't care what orchestrates them.

What you can do today

Spin up an autonomous company from a chat prompt. Tell the CEO "Build a SaaS product with a landing page, email outreach, and social media presence." It hires engineers, marketers, and writers, creates tasks, builds and deploys a real managed runtime app, sets up email inboxes, connects social accounts, and reports back.

Incorporate a real business. Complete KYC, form a US LLC end-to-end ($249, EIN included), and issue virtual Visa cards to AI employees with per-employee spend limits and structured transaction logs.

Automate recurring operations. Set up cron jobs that check email every morning, post to social media daily, run SEO analysis weekly, and generate analytics reports monthly — all via prompts fired to the CEO.

Deploy web applications autonomously. Create managed apps (frontend-only or fullstack with a managed database), deploy previews, promote to production, connect custom domains, and manage environment variables — all from the kanban task system.

Compose primitives into business workflows. Formation → Cards → Domains → Email → Social → Apps. Each primitive feeds into the next. A formed Company owns domains, sends email, uses virtual cards, and posts on social — all through the same API key.

What's next

We're shipping new primitives monthly. The roadmap includes authenticated browser sessions with credential vaulting, insurance provisioning, and deeper integrations with SaaS tools agents need to operate. The goal is simple: every real-world business operation available as an API call.

The platform is live. The docs are comprehensive. The quickstart takes five minutes.

Get started

• Quickstart (5 min): usenaive.ai/docs/getting-started/quickstart
• Full documentation: usenaive.ai/docs
• API reference: usenaive.ai/docs/api-reference/overview
• CLI reference: usenaive.ai/docs/cli/overview
• MCP setup: usenaive.ai/docs/mcp/overview
• Hermes (runtime): hermes-agent.nousresearch.com
• Community: Discord
• Questions? team@usenaive.ai

npm install -g @usenaive-sdk/cli
naive register --name "My Company" --email founder@example.com --password secure123
naive ceo run "Let's build something."