The MachinArc handbook
Understand Machines, connect them into Arcs, and follow work from its first instruction to its final result.
A workspace for AI work
MachinArc is a working prototype for defining, coordinating, and inspecting AI tasks. A Machine holds an objective, a model choice, a set of tools, and execution limits. An Arc connects Machines with decisions, reviews, delays, and an output.
The workspace brings configuration and execution together. You can describe the work, submit a task, inspect its run, and make approval decisions without treating each model response as a separate conversation.
The three building blocks
Start with one Machine and one short task. Build an Arc when the work needs a sequence, a decision, or a handoff. Add automation after the manual path is clear.
- Machine: a configured worker with an objective, tools, permissions, memory settings, and budgets.
- Run: one execution with persisted state, tool activity, approval decisions, usage, and an outcome.
- Arc: a directed workflow that coordinates Machines and operator decisions around a shared input.
What you can evaluate today
The public demo uses Mock Runtime. It lets you explore the product's configuration, workflow, and review experience with simulated model output and external tool behavior. Each visitor receives a separate workspace, credential connections are disabled, and initial schedules are turned off.
Registered workspaces have OpenAI and Anthropic integration paths. A real AI provider acceptance run is still pending because no provider key has been supplied. Demo output demonstrates the product flow; it does not establish the quality or cost of live AI execution.