Question 1

What is Agent Machines?

Accepted Answer

Agent Machines is a per-account agent runtime. Each signed-in user provisions one persistent stateful microVM where the agent's chat history, working files, learned skills, and cron schedules all live on disk at /home/machine and survive every sleep cycle. The same machine wakes on every device the user signs in from.

Question 2

How is this different from a regular chatbot?

Accepted Answer

A regular chatbot is stateless -- conversation memory either lives in your browser or in a vendor-controlled black-box memory service. Agent Machines persists everything to a real Linux filesystem the agent owns: chat .jsonl files, USER.md, MEMORY.md, an FTS5 sessions database, cron schedules, the Python venv, learned skills. The agent can read, write, search, and rebuild context from disk on every wake.

Question 3

Which agents can I run?

Accepted Answer

Two production-ready open-source agents today: Hermes (Nous Research, memory + cron + MCP-native) and OpenClaw (Dedalus Labs reference, Anthropic computer-use loop with browser + screenshot + click). Both expose the same OpenAI-compatible /v1/chat/completions endpoint, both run on the same machine, and you can swap between them from the navbar without losing state.

Question 4

Which providers can host the machine?

Accepted Answer

Dedalus Machines is the default runtime (microVM, second-billed, ~30s cold boot, ~5s warm). The MachineProvider interface also accepts Vercel Sandbox and Fly Machines; users attach their own credentials per provider. Same agent, same persistence, different hosts.

Question 5

How do I sign up and get my own machine?

Accepted Answer

Sign in with Clerk at agent-machines.dev/sign-in. The five-step onboarding (agent / skills / tools / key / boot) provisions the machine end-to-end in 30-90 seconds. Bring a Dedalus API key from dedaluslabs.ai/dashboard/api-keys; the key authenticates both provisioning and inference.

Question 6

What tools and skills come pre-installed?

Accepted Answer

23 built-in tools (terminal, filesystem, browser via Playwright, vision, image generation, MCP servers, code execution, subagent delegation), 17 MCP service integrations (Vercel, Stripe, Supabase, Linear, GitHub, Slack, PostHog, Sentry, Clerk, Firebase, Figma, Shopify, ClickHouse, Datadog, AWS, Cloudflare, plus model providers), and 95 SKILL.md files that auto-load by intent.

Question 7

Where does my data live and who can see it?

Accepted Answer

All conversation, file, memory, and skill data lives on the user's own machine disk under /home/machine/.agent-machines/. The web dashboard's UserConfig (provider keys, machine fleet, active agent) lives in Clerk private metadata, which is server-only and never sent to the browser. No third-party analytics or memory service holds your agent state.

Question 8

What happens when the machine sleeps?

Accepted Answer

Idle Dedalus microVMs hibernate to save cost (second-billed). The disk is preserved across sleeps, so on the next /v1/chat/completions request the machine wakes (typically <5s after recent activity, <30s cold) and the agent resumes from disk -- chat history, learned skills, cron schedules, and venv intact.

Sign in to your fleet