> ## Documentation Index
> Fetch the complete documentation index at: https://vetta.sh/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent

> Naive Managed Agents' core primitive — a versioned config of model, harness, prompt, tools, skills, and budget that runs as durable sessions.

An **agent** is a reusable, versioned *configuration*: a model (and effort), a system prompt, a [toolset](https://vetta.sh/docs/capabilities/tools.md), [skills](https://vetta.sh/docs/capabilities/skills.md), a required [budget](https://vetta.sh/docs/concepts/budgets.md), and a default [completion window](https://vetta.sh/docs/concepts/completion-window.md) — plus the `harness` field that picks the agent loop. Define it once, run it as many [sessions](https://vetta.sh/docs/concepts/sessions.md) — each a durable execution that can pause for hours at storage cost and resume exactly where it left off.

> [card] [Agent reference](https://vetta.sh/docs/concepts/agents.md)
>
> The full agent object, versioning, sessions, harness, completion window, budgets, and policies.

## One agent, many runs

The agent is the *definition*; a [session](https://vetta.sh/docs/concepts/sessions.md) is a single *run* of it. One agent backs many sessions and cron [deployments](https://vetta.sh/docs/capabilities/deployments.md), each independently budgeted, streamed, and controlled.

![One versioned agent definition fanning out to independently budgeted sessions and a cron deployment](https://mintcdn.com/vetta/rjfQyQrWaQ5bFTTq/images/agent-sessions.svg?fit=max\&auto=format\&n=rjfQyQrWaQ5bFTTq\&q=85\&s=c98e8bdc806ddc567a8e6b3c2c41e38b)

## Create one

```bash CLI theme={"system"}
vetta agent create --name Refunder \
  --model z-ai/glm-5.3-flash --harness pi \
  --skill refund-policy \
  --budget-usd 50 --max-task-usd 5 --budget-period month \
  --window immediate --system "You process refunds."
```

```typescript TypeScript theme={"system"}
const agent = await vetta.agents.create({
  name: "Refunder",
  model: "z-ai/glm-5.3-flash",
  harness: "pi",
  skills: ["refund-policy"],
  budget: { capUsd: 50, maxTaskUsd: 5, period: "month" },
  window: "immediate",
  system: "You process refunds.",
});
```

```bash cURL theme={"system"}
curl -fsSL https://api.vetta.sh/v1/agents \
  -H "authorization: Bearer $VETTA_API_KEY" \
  -H "content-type: application/json" \
  -d '{
    "name": "Refunder",
    "model": "z-ai/glm-5.3-flash",
    "harness": "pi",
    "budget": { "cap_usd": 50, "max_task_usd": 5, "period": "month" },
    "window": "immediate",
    "system": "You process refunds."
  }'
```

No computer in the config — sessions provision their own sandbox when needed.

## Fields at a glance

| Field | What it sets |
| - | - |
| `name` · `system` | The agent's name and system prompt. |
| `model` · `effort` | Which model runs each turn, and at what effort. |
| `harness` | The agent loop — `pi` is the default. See [How Naive Managed Agents works](https://vetta.sh/docs/how-managed-agents-is-built.md#the-harness-harness). |
| `tools` · `mcp_servers` | What it may use, each gated `allow` / `ask` / `deny`. See [Tools](https://vetta.sh/docs/capabilities/tools.md). |
| `skills` | Versioned expertise loaded on demand. See [Skills](https://vetta.sh/docs/capabilities/skills.md). |
| `budget` *(required)* | The spend cap every session draws against. See [Budgets](https://vetta.sh/docs/concepts/budgets.md). |
| `window` | The default [completion window](https://vetta.sh/docs/concepts/completion-window.md) — cost vs. latency per request. |
| `output_schema` | A JSON Schema the final result must satisfy. See [Structured outputs](https://vetta.sh/docs/capabilities/structured-outputs.md). |

Every edit mints a new immutable version; the full object is in the [agent reference](https://vetta.sh/docs/concepts/agents.md).

> [card] [Next: computer](https://vetta.sh/docs/capabilities/computer.md)
>
> The disposable micro-VM sandbox, filesystem, shell, and browser the agent works in.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.