Skip to content

Follow your workflows

Open http://localhost:7600 after starting the Muxlio service. If Muxlio runs on another machine, use SSH port forwarding to its loopback address, or the HTTPS address of its authenticated browser gateway.

Find a run

Use Run history to find your workflow and filter by state. Open it to see its current status, inputs, outputs, and any error.

See what is happening

The run view brings the execution graph, attempts, events, logs, and artifacts together. Use Live to follow output while steps run. Open artifacts to inspect the files and reports produced by the workflow.

A successful step has saved outputs. A failed or paused run needs attention: read its error and latest logs before deciding what to do next.

Launch a workflow

Open Workflows to see registered Plan IR versions. Use Run workflow or a row Run action to open the launch dialog. Pick the immutable version and an authorized workspace, then edit the JSON input, which starts at {}. A single workspace is preselected and visible, including the legacy default workspace. If the catalog lists several workspaces, select the one you want to use. For example, if your workstation and the sbx Docker worker both offer Codex, choose localhost for the host checkout or sandbox for the Docker checkout. The workflow can use the same agent/codex executor in either place. Host and sandbox example

Select Preflight to validate the plan, workspace binding, and input schema. The dialog shows the coordinator's diagnostics and unverified checks. A feasible result enables Start run. If you change the version, workspace, or input, run preflight again. When the coordinator returns a run ID, the dashboard opens that run. If a submission response is lost, retry the same version, workspace, and JSON in the same dashboard tab to reuse its submission key.

The dialog requires workspace catalog access. If it reports that your credential cannot read the catalog, ask the operator for workspaces:read access.

Add items to a running workflow

Open a run that declares an appendable collection. Its collection panel shows the accepted items in order and whether the collection is open. Select Add item, enter JSON that matches the displayed Item schema, and choose Submit item. An item with a new ID joins the queue in acceptance order. Select Close collection to finish accepting items and let the queued work drain. These actions require the inputs permission for the run.

For example, the live feature recipe exposes a features collection. Submit {"id":"next-feature","feature":"Describe the next change"} to add work on the current batch branch. That collection closes after 15 minutes without pending work. Docker Sandbox example

Take the next action

Pause or cancel a run when needed. Resume a paused run when it is ready to continue. Retry a failed run after fixing its cause; completed steps are preserved. If an operation's external outcome is uncertain, inspect it under Operations and resolve it with evidence before retrying.

Workers shows which machines and executors are available. Automation manages schedules and webhooks. Retention controls which execution data stays protected from cleanup.

The UI defaults to local operator access. Remote browser hosting requires HTTPS and an authenticated gateway that supplies each user's scoped credential. Read the execution security guide before exposing it remotely. API fields and operational details are in the operations reference.

UI pages use normal paths such as /runs, /runs/<run-id>, and /analytics. These links support direct loading, refresh, and browser back/forward navigation. Existing hash bookmarks are replaced with their path equivalent when opened. Analytics filters and chart selections are preserved in the URL query string.

Analytics shows reported usage, trends, executor activity, and timing/retry breakdowns. Select a chart bar or View attempts to open a titled panel of matching records. Open run opens the run workbench; the step link opens its transcript. Closing the panel or pressing Escape returns to the chart without reloading the dataset. Advanced state, executor, and timezone filters are under More filters. Reported subtotals retain their coverage and do not imply complete consumption when some measurements are missing.