Naive docs

The most efficient managed agent for long-horizon work. The highest intelligence per dollar, by design.

Naive Managed Agents is a managed agent platform: you define an agent — model, prompt, tools, skills, budget — and run it as sessions that work for hours, pause at storage cost, and finish without a person in the loop. You pay per finished task, in real dollars, priced before each call runs.

Two layers, one stack. Naive Managed Agents is the autonomous-company infrastructure — the harness, the managed runtime, sessions, budgets, tools. Naive blueprints are the layer on top: the templates a whole company is installed from, and the studio you run it in. These pages are the Managed Agents reference.

Why you’ll see vetta. Naive’s API, SDK and CLI use the name vetta in their identifiers: api.vetta.sh, the Vetta-Version header, @usenaive-sdk/vetta, @usenaive-sdk/vetta-cli and the vetta command. They’re the same product. The CLI also installs a naive command that runs the same commands (naive login is vetta login).

Everything is built around one number: the most intelligence per dollar spent — the same model on different stacks varies several-fold in cost per completed task (benchmarks).

Any model, your skills, and your tools go in; Managed Agents’ runtime does the work; and you run and interact through the Web Studio, the API, the CLI, or the SDK.

​

Create an agent, run a session

TypeScript

import { createClient } from "@usenaive-sdk/vetta";

const vetta = createClient({ baseUrl: "https://api.vetta.sh", apiKey: process.env.VETTA_API_KEY! });

const agent = await vetta.agents.create({
  name: "triage",
  model: "z-ai/glm-5.3-flash",
  system: "You are a concise triage agent.",
  budget: { cap_micro_usd: 50_000_000, max_task_micro_usd: 5_000_000, period: "month" },
});

const session = await vetta.sessions.create({
  agent_id: agent.id,
  message: "Summarize what changed in main overnight.",
});

for await (const event of vetta.stream(session.id)) {
  console.log(event.type);
}

​

The pillars, as objects

Everything you touch maps to primitives you create and version — familiar on day one.

Agent

A reusable, versioned config: model, harness, prompt, tools, skills, budget.

Session

One directly-controlled run of an agent — send input, stream output, interrupt, resume.

Events

A resumable stream of everything the agent does, keyed by sequence number.

Computer

A disposable sandbox with a filesystem, shell, and optional browser — paused when idle.

Skills

Persisted, versioned expertise loaded with progressive disclosure.

Files

Durable, organization-scoped storage for the artifacts a session produces.

​

How Naive Managed Agents earns the highest ROI per dollar

Four independent levers, all cutting dollars per finished task without cutting quality.

​

Available today & what’s next

Available now: agents, sessions, the durable runtime, completion windows, budgets, the computer, skills, files, cron, webhooks, real-USD billing, identity (personas, domains, email, SMS, connections, the vault) — plus web search and fetch, image and video generation, the Messages-format model proxy, observability, structured outputs, audit logs, and teams — a coordinator with a version-pinned roster, delegation, and a shared board.

Quickstart

Create an agent, run it, and read the meter in under five minutes.

How Managed Agents works

Harness, runtime, tools — what each layer is, and which one you pick.