Developers

Import & export

Export downloads the graph as a madmax/v1 document with secrets stripped. Import takes a template, a zip folder export, or an n8n export with placeholders.

Workflows are portable JSON documents.

Export

Canvas menu (⋮) → Export downloads <workflow-name>.madmax.json, the graph in the madmax/v1 template format. Connection bindings, webhook ids, and anything workspace-local are stripped: secrets never travel in an export, so the file is safe to commit, share, or attach to a ticket. The same document comes from GET /api/v1/workflows/:id/export.

Import

On the workflows index, the New workflow modal takes a file (or pasted JSON): a MadMax template, a .zip folder export, or an n8n workflow export, up to 20 MB.

The import dropzone on the new-workflow modal Drop a .madmax.json or an n8n export; the importer figures out which.

Importing a MadMax template recreates the graph; you re-bind connections (the AI's auto-wire rules apply) and re-verify.

What travels with a workflow

Secrets and connections never do. The data a workflow is attached to does:

  • Data stores a store_get / store_upsert step uses are referenced by slug, and the export carries each store's definition (name, schema, row cap), never its rows. Rows are your data; like secrets, they stay in the workspace.
  • Memories an agent step binds travel by name with their settings (window, compaction, recall).
  • Knowledge bases travel by name and are created empty; re-upload or re-index the content.

On import, anything missing is created in the target workspace before the graph lands; an existing store with the same slug keeps its schema.

n8n imports

An n8n export is translated, and the translation never fails:

  • Nodes with a MadMax equivalent map directly: HTTP request, if/switch (cases become ports), loops (batches become iterate/done), set, code, and the trigger types.
  • Nodes without an equivalent arrive as clearly-annotated placeholder steps that pass data through unchanged, so the shape of your workflow survives even where the pieces don't.
  • The import report lists everything that needs attention, and the placeholders seed a ready-made prompt: ask Max to finish the build, and it replaces placeholders with real MadMax steps.

Sticky notes and n8n-specific AI sub-nodes are dropped (noted in the report).

‹ Webhooks Workspaces & members ›