Blueprints
Browse published workflow and agent blueprints and grab them into your project as independent, editable copies
What Blueprints are
A blueprint is a ready-made workflow or agent, published to a catalog that every organization on the platform can browse. When you find one you want, you grab it into one of your projects, and what lands there is a brand-new, fully independent copy that you own and can edit freely.
Blueprints exist so a good build doesn't stay trapped in the project it was built in. A KYB verification flow, a document-compliance reviewer, a support-triage agent: someone builds it once, publishes it, and every other team starts from a working version instead of a blank canvas.
Blueprints are an in-product catalog of templates. This is different from Cloud Marketplace, which is about buying a MagOneAI subscription through AWS, Azure, or GCP.
The model in one paragraph
A blueprint stores an immutable, sanitized snapshot of the source asset, completely decoupled from it. Publishing exports the source's Hub-published version, strips anything sensitive or environment-specific, and stores the result. Grabbing copies that stored snapshot into your project as a new asset with fresh identifiers and no link back. Editing or re-publishing the source workflow never changes the listing; removing or republishing the listing never changes anything anyone has already grabbed.
Browsing
Blueprints appears in your organization's navigation. The listing page shows every published blueprint, with:
- Search across blueprints
- Category filter: Productivity, Sales, Support, Finance, Engineering, HR, Marketing, Other
- Type filter: All, Use Cases, or Agents
- Sort: Newest first (default), Oldest first, Most grabbed, or Name (A to Z)
Results are paginated at 12 per page.
The detail page
Opening a blueprint shows its full description, which supports rich formatting, embedded images, video, and downloadable documents, alongside a details panel:
| Field | Meaning |
|---|---|
| Type | Use Case or Agent |
| Category | Its catalog category |
| License | MIT, Apache 2.0, Proprietary, or Custom |
| Grabs | How many times it has been installed |
| Published | When the listing was created |
| Dependencies | External integrations it needs, plus notes about what isn't included |
Two actions are always available: Grab it, and Preview, a read-only view of the actual workflow or agent configuration, so you can see how it's built before committing.
Grabbing a blueprint
Review what it needs
The grab dialog opens with the full description and a What this needs panel listing the external integrations the blueprint expects, or telling you plainly that no external integrations are required.
Choose a project
Pick the project to grab into. You need write access to that project.
Name it
The blueprint's name is pre-filled; change it to whatever makes sense in your project. If a workflow with that name already exists there, the grab is refused with a clear conflict message. Rename and try again.
Check the readiness report
Once a project is selected, the dialog re-checks the dependencies against that project and marks each integration either connected or configure after grab, so you know what's left to wire up before the first run.
Confirm
Confirm the grab. On success you can jump straight to it: Open in Canvas for a use case, Go to Agent for an agent.
You can also start from a blueprint when creating a use case: the create panel has a From blueprint mode alongside Blank, with its own searchable picker.
What you get, and what you have to finish
A grabbed blueprint is a genuine copy: new identifiers throughout, no foreign key back to the listing, no shared state with anyone else who grabbed it. Edit it however you like.
To make that copy portable and safe, some things are deliberately not carried over:
| Not included | What to do |
|---|---|
| Knowledge bases | Attach your own after grabbing. The dependency notes say so explicitly. |
| LLM configurations | Your project's default model is used unless you set one. |
| MCP connections and connection hints | Connect the required integrations in your project. |
This is why the readiness check matters. A blueprint that needs an integration will import cleanly and then fail at run time if that integration isn't connected. The panel tells you which ones up front.
For a use case, the grab brings across the whole tree: the workflow itself, the agents it references, every child workflow it calls, and (for a Super Agent) the use cases it can call as capabilities. All of them are recreated in your project.
Publishing a blueprint
Publishing is a superadmin action, done from Studio against a real asset in a real project.
Publish the source to the Hub first
Only a Hub-published version can be snapshotted. The whole point is that a blueprint ships something known-good, not a work in progress. If the workflow calls child workflows that aren't published, publishing is blocked with the exact list: "Publish these child workflows to the Hub first: …"
Open the publish dialog
On a use case card, the menu offers Publish as Blueprint (or Republish Blueprint if it's already listed). Agents are published from the agent card's publish action. Only library agents can be published.
Write the listing
Fill in the catalog metadata:
- Name (required): up to 255 characters
- Short description (required): one line shown on listing cards, up to 160 characters
- Category (required)
- License: MIT, Apache 2.0, Proprietary, or Custom
- Long description (optional): full Markdown with a live preview, plus uploaded images, documents, and video, or an embedded YouTube / Vimeo / Loom URL
Publish
The snapshot is exported, sanitized, checked against the publish guards, and stored. From that moment the listing is frozen: editing the source workflow changes nothing until you explicitly republish.
What sanitizing removes
Before a snapshot is stored, the same strippers that run on any workflow export are applied, removing knowledge base ids, LLM configuration ids, MCP connection ids, and connection hints, and the result is scanned for secret patterns. For a Super Agent, the per-node list of disabled capabilities is stripped too, so the listing carries the workflow's structure, never one project's environment.
Publish guards
A publish is refused if any of these are true:
| Guard | Limit |
|---|---|
| A secret or API-key pattern is detected in the snapshot | Blocked, naming the field |
| The exported snapshot is too large | 5 MB |
| The workflow has too many activities | 100 |
| An identical snapshot is already listed for this organization | Blocked as duplicate content |
| A referenced child workflow isn't Hub-published | Blocked, naming the workflows |
| The agent isn't a library agent | Blocked |
The duplicate check hashes the content itself, ignoring the export timestamp. That's what makes a republish with no real changes a no-op instead of a second identical listing.
Republishing, editing, and removing
| Action | Effect |
|---|---|
| Republish | Refreshes the snapshot from the current source. The listing's link stays the same, so anyone who bookmarked or shared it still lands in the right place. |
| Edit details | Changes display metadata only: name, descriptions, category, license, media. No re-snapshot, and it works even if the source asset is long gone. |
| Delete | Soft-deletes the listing so it disappears from the catalog. Existing grabs are unaffected, because they're independent copies. |
A Super Agent blueprint's content hash depends on the whole project's published state, because its callable capabilities are part of what gets bundled. Publishing an unrelated workflow in the same project can therefore change the hash, which means a republish that looks like a no-op may not be one.
Media in a listing
Long descriptions carry more than text, because a blueprint often needs a walkthrough:
| Kind | Formats | Size limit | Behaviour |
|---|---|---|---|
| Image | PNG, JPG, WebP | 2 MB | Rendered inline |
| Document | PDF, DOC/DOCX, XLS/XLSX, CSV, PPT/PPTX, TXT, ZIP | 25 MB | Downloads when clicked |
| Video | MP4, WebM, MOV | 200 MB | Streamed with seek support |
Hosted videos need no upload at all. Paste a YouTube, Vimeo, or Loom URL on its own line and it becomes a sandboxed embed.
Uploads are superadmin-only and validated by inspecting the file's actual bytes rather than trusting the declared type; SVG, HTML, and executables are rejected outright. Because a blueprint's images and video have to load for viewers in every organization, the files themselves are served publicly, so don't put anything confidential in listing media.
Who can do what
| Action | Required access |
|---|---|
| Browse listings, view detail, preview | Any signed-in user |
| Preview a grab and grab into a project | Write access to the target project |
| Publish, republish, edit details, delete, upload media | Superadmin |
Limitations today
- Public visibility only. Every published blueprint is visible platform-wide. Organization-only listings are reserved for a later phase.
- Publishing is superadmin-only. Organization-level publishing isn't available yet.
- No moderation queue. There's no review or rejection workflow; a published listing is live immediately.
- Snapshots don't auto-update. A blueprint reflects the source as of its last publish or republish, by design.
Next steps
Versions and publishing
How Hub publishing works, the prerequisite for publishing a blueprint
MagOne Studio
Where blueprints are published from and grabbed workflows are edited
Tools and integrations
Connect the integrations a grabbed blueprint expects
Knowledge bases
Attach your own knowledge after grabbing, since they're never bundled