The development loop

Try it in the REPL, edit locally, sync to the cloud, run on your instance, read the result, repeat.

Three copies of a project

CopyWhereChanged by
Your working treeA folder on your computerYou and your coding agent, with any editor
The cloud projectThe Zeon cloudzeon push / zeon sync, and saves in the web app (World Builder, Skills Editor, Object Database)
The loaded copyYour cloud instance (or a lab system)Loading the project: zeon up, a refresh after zeon push / zeon sync, or opening the project in the web app

Your instance runs the loaded copy, never your folder. A change reaches a run only after it is pushed and loaded.

The loop

  1. Try in the REPL. zeon py '<code>' runs Python in the simulation on your instance at once: move an arm to an anchor, read the world, call a skill with other inputs (skills.<id>(...)), look at the result in the viewer or with zeon snap. Adjust and try again until it works. See The REPL.
  2. Edit skills, workflows, worlds and canvases in your working tree: write in what worked.
  3. Sync. zeon sync commits what you changed, merges what changed in the cloud, and pushes. When your instance has this project loaded, it then reloads it and prints instance refreshed (the loaded world is kept; --world W also reloads a world). No instance refreshed in the output: run zeon up.
  4. Run in the simulation: zeon run NAME from the terminal, or the run screen in the web app (zeon app NAME --run prints a signed-in link). zeon run runs Fast at 8x by default: every move planned as on the robot, then played back (--speed 1|2|4|8). Iterate that way; before the real robot, run once with --motion full (the robot's control loops, in real time: add --detach, then zeon wait ID).
  5. Read the result: zeon logs ID, the Log tab in the app, the 3D view. A failed run's error names the skill, the step and the line of its code; the next: line says what to do. When a run stopped half-way with an arm holding or touching something, zeon reset-arms puts both arms back at their start position before the next run.
  6. Repeat from 1.

When you are done for the day, zeon down releases your instance.

# once
zeon clone my-project && cd my-project
zeon up

# try things
zeon brief
zeon py 'skills.move_plate(destination="wellplate_stand_2", destination_anchor="top_center")'

# every change
zeon sync
zeon run plate_transfer -i plate=wellplate
zeon logs <id>

When the app changed the project

Saving in the web app (saving a world in the World Builder, saving an object's anchors in the Object Database) writes a new commit to the cloud project. Before you edit the same files locally, bring that commit in:

zeon pull     # refuses while you have uncommitted changes: commit first

or simply zeon sync, which merges both sides. If both sides changed the same lines of a file, the sync stops and asks you to resolve the conflict. See Conflicts and a diverged project.

Notices from the CLI

zeon run, zeon py and zeon brief compare your folder with what your instance runs and print notice: lines on stderr when they differ, each naming the command that fixes it:

Notice saysRun
your local edits are not pushedzeon sync
your local commit is not pushedzeon push
the instance has another project (or none) loadedzeon up
the instance runs an older commit than your last pushzeon up
the cloud project moved on (for example, a save in the app)zeon pull or zeon sync

Act on a notice before reading a failure as a bug in your code.

Simulation first, then real arms

Your cloud instance only simulates. A clean simulated run shows that the plan is collision-free against the models; it does not measure the real cell. Run on real arms from a lab system, which loads the same cloud project. See Running a workflow on real hardware.


Did this page help you?