5. Autonomous Agents¶

Backend.AI GO includes an Autonomous Agents platform that lets you run AI agents on your local machine in a fully isolated environment.

Providers¶

Hermes is the recommended provider for new setups. It runs entirely inside a container, so your host system remains clean and recoverable if anything goes wrong.

Runtime support (desktop vs headless)¶

Autonomous Agents providers are available only in the desktop app. Each provider drives a local container gateway and attaches a native desktop event bridge, neither of which the standalone headless server hosts.

When the WebUI runs against a headless server, the Autonomous Agents page reports each provider as unavailable with an explicit reason instead of presenting setup or execution flows that cannot work. Two surfaces make this explicit:

• GET /api/v1/autonomous/availability returns a per-provider availability report plus a feature-level empty-state reason. A provider is available only when it is registered in the running process; the headless server registers no desktop-only providers, so each entry reports available: false with an unavailableReason.
• GET /api/v1/autonomous/providers continues to list only registered providers, so it returns an empty array on the headless server, consistent with the availability report.

A future provider that does not need the desktop runtime would be registered in the headless server too and would then report available: true automatically. To set up and operate the current providers, run the desktop app.

Prerequisites¶

Before using Autonomous Agents you need a supported container runtime:

• Docker Desktop (macOS, Windows, Linux) — most common choice
• Apple Container (macOS arm64 only) — lighter-weight alternative using Apple's Containerization framework

Getting started¶

• Setting up Hermes — step-by-step onboarding wizard guide