Key concepts
The vocabulary used throughout Zeon: project, skill, workflow, canvas, world, object, anchor, setup, sim and real, cloud instance, REPL, sync.
Project
A project is a folder of plain text files, versioned in the Zeon cloud. Its layout:
| Path | What it holds |
|---|---|
project.json | The manifest: name, description, active_workflow, active_world, and setup (the rig the project runs on). |
skills/<id>/ | One skill: robotic_code.py (the code), metadata.yaml, modules.py (imports the skill API). |
workflows/<id>.json | One workflow graph per file. |
worlds/<id>/ | One world: world_state.json (the scene). |
objects/<type>/ | One object type: <type>.urdf and <type>.object_model.yaml (parts, articulations, anchors). |
canvas/<id>.tsx | Optional input screens for workflows. |
viewer.yaml | Optional camera presets for the 3D viewer. |
AGENTS.md, CLAUDE.md | The guide for coding agents. |
You keep a local copy (a working tree) and move changes to and from the cloud with zeon sync. See Syncing your work.
Skill
A skill is one robot action written in Python, such as "move a plate from one seat to another". Skills call the Zeon skill API: modules.py imports it and robotic_code.py imports what it uses from .modules. A skill's inputs are the parameters of its Python function. A skill fails by raising an exception. See Authoring a skill and the skill API.
Workflow
A workflow is a graph of nodes stored as JSON: a start node, skill nodes, and an end node, connected by edges that say when to move on (for example on success). A workflow declares inputs with default values; skill-node parameters bind to them. See Authoring a workflow.
Canvas
A canvas is an optional React screen, canvas/<id>.tsx, that collects a workflow's inputs before a run. The workflow's canvas_ui block attaches it. You or your coding agent write the file; the web app renders it on the run screen. See Canvases.
World and object
A world is a scene: which objects are in the cell and where they sit, including the tables. Each object in a world has an instance id (for example wellplate_stand_1) and a type (for example wellplate_stand). When a world loads, each type is taken from the project's objects/<type>/ folder, or else from the shared mesh database catalog your organization can see. See Building a world and The mesh database.
Anchor
An anchor is a named frame on an object, defined in its object model: where the gripper grasps it (grasp_short), where another object sits on it (top_center, slot_1), and so on. Skills move to anchors instead of to coordinates written in code, so re-teaching an anchor changes the motion without a code change. See Anchors.
Setup
A setup describes a robot rig: its arms, their models and grippers, and their calibration. A project names the setup it is written for in the setup field of project.json. When you load a project on your cloud instance, the instance switches to that setup if your organization may use it. A lab system runs the setup it was installed with.
Sim and real
A run is either simulated or real. In the simulation the arms and objects are models; nothing moves physically. Your cloud instance runs simulations only. Real runs happen on a lab system attached to the arms.
Cloud instance
Your account is bound to one cloud instance while you work. Both the web app and the CLI use it. zeon up loads your project on it; zeon down releases it. When every instance is in use, you wait in a queue: the web app shows your place in line and the CLI reports that no instance is free.
REPL
The REPL is a live Python session on your cloud instance, in the simulation. You send code with zeon py; it runs at once against the loaded project and world, and you see the result. Names you define stay defined between entries. It is the main way to try moves and skills before you write them into files. See The REPL.
Sync
Sync moves files between your working tree and the cloud project. It works like a small version-control system: zeon commit records a snapshot locally, zeon push sends it, zeon pull brings cloud changes in, and zeon sync does all three. The cloud project is what your cloud instance and lab systems load. See Syncing your work.
Run
A run (an execution) is one execution of a skill or a workflow. Each run has an id. You start runs from the web app or with zeon run, and read their step-by-step log in the app or with zeon logs.
Updated 8 days ago