> For the complete documentation index, see [llms.txt](https://docs.gotempest.app/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.gotempest.app/ai-and-automation/external-agents-vs-llm-providers.md).

# External Agents vs LLM Providers

Compare External Agents and LLM Providers in Tempest: terminal tools, permissions, attachments, model settings, and setup requirements.

Tempest offers two ways to work with AI in its chat panel. **LLM Providers** supply a model to Tempest's built-in Storm AI workflow. **External Agents** run a separate agent through ACP (Agent Client Protocol), with that agent managing the conversation and tool use.

Choose an **LLM Provider** for Storm AI's built-in tools and permission controls. Choose an **External Agent** to bring an agent you already use, with its own models, settings, and workflow.

## Capability comparison

| Capability                                     | LLM Providers                                                                                                                                                    | External Agents (ACP)                                                                                                           |
| ---------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------- |
| Examples                                       | Built-in Storm AI, custom Anthropic/OpenAI/Gemini endpoints, OpenAI-compatible local models, Apple Intelligence.                                                 | Claude Agent, Codex, OpenCode, Cursor, Qwen Code, or a custom ACP agent.                                                        |
| Chat inside Tempest                            | Yes.                                                                                                                                                             | Yes.                                                                                                                            |
| Explain output and suggest commands            | Yes.                                                                                                                                                             | Yes.                                                                                                                            |
| Execute commands and read live terminal output | Yes, through Tempest's tools when the selected permission mode allows it.                                                                                        | Yes, through the Tempest tool bridge and injected MCP server.                                                                   |
| Interact with terminal programs such as `htop` | Yes, using terminal screen and input tools.                                                                                                                      | Yes, using Tempest's terminal tools.                                                                                            |
| Who controls the tool workflow?                | Tempest runs the tool loop and exposes a controlled tool set to the model.                                                                                       | The external agent runs its own tool loop and uses its own tools plus Tempest MCP.                                              |
| Permission modes                               | Tempest's Read Only, Ask Permissions, and Bypass modes, subject to policy.                                                                                       | Modes advertised by the agent, such as its approval or sandbox settings. Tempest's LLM permission-mode selector does not apply. |
| Approval prompts                               | Tempest checks tool calls and asks according to the selected mode; third-party MCP calls require approval.                                                       | ACP permission requests appear in Tempest's approval UI. Which calls request approval depends on the agent and its settings.    |
| Model and reasoning settings                   | Select a configured model in Tempest; settings depend on the provider.                                                                                           | Tempest displays model, reasoning, and other options when the agent exposes them through ACP.                                   |
| File and document attachments                  | Yes; Tempest extracts supported document formats locally before sending text.                                                                                    | Yes; attachments are sent as ACP resources, links, or text according to the agent's capabilities.                               |
| Image attachments                              | Depends on the selected model. Apple Intelligence is text-only in Tempest.                                                                                       | Depends on the agent's advertised image support and selected model.                                                             |
| Tempest Drive documents and skills             | Built-in tools can read Drive documents and load/save skills.                                                                                                    | Selected Drive documents can be attached to the conversation; the agent's own skills and tool support depend on that agent.     |
| Additional MCP tools                           | Import MCP servers in Tempest's AI settings; Storm AI applies its approval controls.                                                                             | Tempest injects its own MCP server. Configure other MCP servers through the agent's supported configuration.                    |
| Web search                                     | The built-in Storm AI provider offers a web-search option, subject to plan and policy. Custom providers and Apple Intelligence do not have that built-in toggle. | Depends on the agent's tools; Tempest's Storm AI web-search toggle does not apply.                                              |
| Local installation                             | No agent executable needed. Custom providers need endpoint/model settings; Apple Intelligence needs a supported Apple device.                                    | Install the preset's runtime or CLI on the machine running Tempest.                                                             |
| Authentication                                 | Built-in Storm AI uses your Tempest account. Custom API providers use their configured keys; Apple Intelligence needs no API key.                                | Uses the agent's own sign-in or provider credentials.                                                                           |

Features remain subject to your Tempest plan and team policy. An agent or model's capabilities can differ by version and configuration; a preset does not guarantee every optional feature.

## Get started

* [Use Storm AI in Tempest](https://docs.gotempest.app/ai-and-automation/tempest-ai-assistant.md) — use the built-in workflow or configure an LLM provider.
* [Use AI Agents in Tempest](https://docs.gotempest.app/ai-and-automation/using-ai-agents-to-manage-servers.md) — ACP presets, requirements, and setup.
* [Storm AI Security Model](https://docs.gotempest.app/ai-and-automation/tempest-ai-security-model.md) — tool access and approval behavior.

This page compares the two choices **inside Tempest**. To work in an external AI client and give it access to Tempest, see [Use Tempest in AI Agents](https://docs.gotempest.app/ai-and-automation/install-tempest-mcp-server-in-ai-clients.md), which explains Desktop and CLI MCP modes.
