---
title: "FAQ & limits"
section: "Autonomous Agents"
description: "Answers about SmartestQA autonomous agents: changes, cost, Claude Desktop, permissions, data sent to AI, agents versus MCP, and every limit."
url: https://docs.smartestqa.ai/en/ai-agents-faq.html
audience: [customer, admin]
updated: 2026-11-08
keywords: "agents faq cost budget claude desktop permissions data ai provider mcp limits max steps daily budget"
---
# Autonomous Agents: FAQ & limits

Answers to common questions about autonomous agents, how agents relate to AI assistants, and the module’s limits.

## In short

**On this page you can**

- Get short answers about changes, cost, permissions and data.
- Filter the questions for customers or admins.
- Run a five-minute agent demo for a prospect.
- Look up every agent limit in one table.

**How to use it**

1. Pick your audience in the filter bar above the questions.
2. Open a question; follow its link for the full procedure.
3. Check Limits before you change an agent’s settings.

## Questions and answers

### Q: Will an agent change my tests on its own?

No. When the SmartestQA server runs an agent, every change is recorded as a proposal and waits until a person clicks **Apply**. New agents also start paused. See [Apply or dismiss proposals](https://docs.smartestqa.ai/en/ai-agents-runs.html#proposals).

_Audience: customer, admin_

### Q: What does an agent cost to run?

A server run uses tokens of the [project’s AI connection](https://docs.smartestqa.ai/en/project-settings-ai.html#ai-account); the cost of each run is shown in **[Runs](https://docs.smartestqa.ai/en/ai-agents-runs.html#runs-tab)** and capped by **[Max cost per run](https://docs.smartestqa.ai/en/ai-agents-settings.html#limits)** and **Daily budget**. A run in Claude Desktop runs on your Claude plan and records no cost in SmartestQA. See [How spend is counted](https://docs.smartestqa.ai/en/ai-agents-numbers.html#spend).

_Audience: customer, admin_

### Q: Do I need Claude Desktop?

No. The report agents (Daily triage, Daily QA report, Weekly management report, Coverage gap report) run only on the SmartestQA server, and you can set any other agent to run there. [Claude Desktop](https://docs.smartestqa.ai/en/mcp-connect.html#claude-desktop) is an option for agents that work through [MCP tools](https://docs.smartestqa.ai/en/mcp-tools.html). See [Runs on](https://docs.smartestqa.ai/en/ai-agents-settings.html#runs-on).

_Audience: customer_

### Q: Does an agent run when my computer is off?

On the server, yes. In Claude Desktop, no: Claude Desktop must be open at the scheduled time. If Claude Desktop asks later in the same period, the run still happens, unless its [**Late cutoff (HH:mm)**](https://docs.smartestqa.ai/en/ai-agents-settings.html#claude-task) has passed.

_Audience: customer_

### Q: Where do the reports go?

Always into the [run detail](https://docs.smartestqa.ai/en/ai-agents-runs.html#run-detail), with **Download PDF**. Optionally to Slack, Microsoft Teams, Discord or WhatsApp (from [Integrations](https://docs.smartestqa.ai/en/integrations-notifications.html)), up to 10 e-mail addresses (formatted, with a PDF) and a webhook. See [Reports](https://docs.smartestqa.ai/en/ai-agents-settings.html#reports).

_Audience: customer_

### Q: Which agents should I install first?

The ones marked **Core**: Daily triage, Daily QA report and Coverage gap report, and the two [deploy agents](https://docs.smartestqa.ai/en/ai-agents-catalog.html#on-deploy) once your [CI reports deploys](https://docs.smartestqa.ai/en/deployments-ci.html). Try each on-demand agent once. See [The catalog](https://docs.smartestqa.ai/en/ai-agents-catalog.html#which-agent).

_Audience: customer_

### Q: What is the difference between an agent and an AI assistant through MCP?

An agent works alone on a rhythm and leaves you a report and proposals. An assistant through [MCP](https://docs.smartestqa.ai/en/mcp.html) works with you in a chat: it [shows you each change as a diff](https://docs.smartestqa.ai/en/mcp-safe-changes.html#propose-apply) and writes only after you say yes. Both use your SmartestQA [permissions](https://docs.smartestqa.ai/en/company-settings-roles.html#matrix).

_Audience: customer_

### Q: Who can create, run and delete agents?

Every [role](https://docs.smartestqa.ai/en/company-settings-roles.html#matrix) (Super Admin, Admin, User) in the projects they can open. Tools that need a permission (such as `VIEW_REPORT`) are refused for a user whose role lacks it; admins have every permission. Today the User role has no permissions assigned, and [roles are fixed](https://docs.smartestqa.ai/en/company-settings-roles.html), so an Admin cannot add one. [Deleting an agent](https://docs.smartestqa.ai/en/ai-agents-catalog.html#pause-delete) keeps its run history.

_Audience: admin_

### Q: Whose permissions does an agent use?

A scheduled or deploy run uses the agent’s creator. [**Run now**](https://docs.smartestqa.ai/en/ai-agents-runs.html#run-now) uses the person who clicked it. An applied proposal runs as the person who applied it. Each can reach only the [projects that person can reach](https://docs.smartestqa.ai/en/company-settings-users.html#project-access).

_Audience: admin_

### Q: What data is sent to the AI provider?

For a server run: the agent’s instructions and the results of the tools it calls (for example [run results](https://docs.smartestqa.ai/en/report.html), test case names, [defects](https://docs.smartestqa.ai/en/tasks.html), [app map](https://docs.smartestqa.ai/en/discover-app-map.html) pages) go to the [AI provider configured for the project](https://docs.smartestqa.ai/en/project-settings-ai.html#ai-account). Each model call is logged with its tokens and cost, under the query type `AI_AGENT`. For a Claude Desktop run the data goes to Claude through your [MCP connection](https://docs.smartestqa.ai/en/mcp-connect.html).

_Audience: admin_

### Q: Is every run kept?

Yes: status, every step with its input and output, tokens, cost and the settings used. The **[Runs](https://docs.smartestqa.ai/en/ai-agents-runs.html#runs-tab)** tab shows the latest 100 of the project; older runs stay stored.

_Audience: admin_

### Q: Can my AI assistant install agents for me?

The assistant’s [status overview](https://docs.smartestqa.ai/en/mcp-tools.html#tools-overview) (`smartest_status`) tells you when no agent watches the project (the [work key](https://docs.smartestqa.ai/en/mcp-tools.html#work-keys) `AGENTS`). [Installing and switching agents on](https://docs.smartestqa.ai/en/ai-agents-catalog.html#install) is done on this page.

_Audience: customer_

## Limits

| Item | Limit |
| --- | --- |
| Agent name · description | 150 · 1,000 characters |
| [Max steps per run](https://docs.smartestqa.ai/en/ai-agents-settings.html#limits) | 1–60 |
| Max cost per run | above 0, up to 10 USD |
| Daily budget | above 0, up to 100 USD |
| E-mail recipients per agent | 10 |
| Runs of one agent at the same time | 1 |
| Runs listed on the Runs tab | latest 100 |
| Bars in Last runs | 14 |
| Deploy check for [After each deploy](https://docs.smartestqa.ai/en/ai-agents-settings.html#deploy-trigger) | every hour |
| Deploy events handed to one run | 20 (the rest go to the next run) |
| [Overdue warning](https://docs.smartestqa.ai/en/ai-agents-numbers.html#overdue) (Claude Desktop) | 90 minutes after the scheduled time |
| [Stuck run closed](https://docs.smartestqa.ai/en/ai-agents-numbers.html#stale) | Queued 15 min · Running 10 min · Claude Desktop 6 h |
| Catalog agents per project | 1 per catalog entry |
