Executions panel on the canvas, Super Agent contracts, and usage limit requests

This release brings your workflow's run history onto the canvas, makes Super Agents more predictable, and adds a simple way for members to ask for more usage. Tool connections also become more secure and stay signed in longer, and the canvas is easier to work with.

Highlights

Executions panel on the canvas

An execution is a single run of a workflow. A new executions panel on the canvas in MagOneAI Studio lists every run of the workflow you're editing. For each run it shows the status, what started it (for example a person, a schedule or the API), how long it took, how many AI tokens it used, and who ran it. From the panel you can open a run to inspect it, cancel one that's still going, or re-run it.

Why it matters: You can build a workflow and check how it behaves in one place, without switching back and forth to the Executions page.

Learn more: Triggers & execution

Super Agent inputs and outputs

A Super Agent is a node that plans its own work and hands parts of it to your project's published workflows. You can now declare exactly which inputs a Super Agent receives and which outputs it returns, and map them in the properties panel. MagOneAI enforces this contract on every run.

Files are also handled better. When a Super Agent calls one of its workflows, the files that workflow needs are now passed along automatically instead of relying on the AI model to do it.

Why it matters: A Super Agent now fits into a larger workflow like any other step, always receiving and returning what the rest of the workflow expects.

Learn more: Super Agent node

Requests for higher usage limits

Organizations share a usage allowance, measured in AI tokens. Members who are running low can now request a higher limit, and an admin approves or declines the request. A tabbed Usage page shows how much of the shared allowance has been used. Approved top-ups are temporary and expire after a set period.

Why it matters: Nobody gets stuck when they hit a limit, and admins keep control of spending.

Learn more: Usage & quotas

Conversation attachments

The Conversations page and the chat run panel now show the files shared in a conversation, with thumbnails, together with the files the agent produced and the sources (citations) it used in its answers.

Why it matters: Everything about a conversation, including what went in, what came out and where the answers came from, is in one place.

Tool connections that ask only for what they need

Tools let agents act in other systems, such as Jira or Outlook. Many connect using OAuth, the familiar "Sign in with…" flow where you grant MagOneAI permission to act on your behalf. A connection now asks only for the permissions needed by the tools you actually selected, and agents only see those tools.

Why it matters: You grant the least access necessary, which is safer and easier to approve for IT and security teams.

"Used in" view for child workflows

A workflow can call another workflow as one of its steps, which we call a child workflow. The use case page now has a Used in view listing every workflow that runs this one as a child.

Why it matters: Before you change a workflow, you can see everything else that depends on it.

Canvas settings

New canvas settings let you show or hide how data flows between steps, and switch on trackpad-style navigation for panning and zooming.

Extra security for OAuth connections (optional)

Connections to services that use dynamic OAuth sign-in can now turn on PKCE, an additional security check that protects the sign-in exchange from being intercepted.

Improvements

• Connections stay signed in. Sign-ins for connected tools now refresh automatically, so you no longer have to reconnect. Atlassian connections no longer drop after the first hour.
• Child workflows always use a published version. Only published workflows can be chosen as child workflows, and a preview shows exactly what you're picking. The Use Case and ForEach nodes record which published version ran. A ForEach node pointing at an unpublished workflow now fails with a clear error. See Versions & publishing.
• Better recovery when an AI provider is busy. When a model provider limits requests or runs out of quota, runs keep retrying for longer and explain the problem clearly instead of simply failing.
• Setup guidance before you connect. The connect panel lists any setup an OAuth app needs in the provider's console before you sign in (for example, enabling Atlassian APIs). The Jira and Confluence buttons now name the right provider.
• An easier-to-scan node palette. Building blocks are grouped into collapsible Agentic, Branching, Tools and System sections, and the canvas toolbar is cleaner.
• A richer execution timeline. Each step shows a usage summary, and runs of child workflows are labeled with the workflow's name and item number.
• A tidier START node editor. The START node defines a workflow's inputs. New fields appear in the right order, field keys stay unique when you reload, and durations show in hours.
• Wiring mistakes caught before a run. Validation now catches broken connections between steps' inputs and outputs before a workflow runs, instead of halfway through.
• Pasted keys are cleaned up. Credentials for tool servers are trimmed, so a key pasted with an accidental trailing line break no longer breaks the connection.

Bug fixes

• Secrets removed from exports. API credentials in blueprints and exported workflows are replaced with a placeholder, so exports no longer contain real passwords or keys and still import cleanly.
• Open browser tabs no longer cause outages. We fixed outages, including failed sign-ins, caused by many open browser tabs holding database connections for notifications.
• No runs on failed uploads. Running a workflow, starting a chat, saving a schedule and running an agent are now blocked if a file upload failed or produced an empty transcript.
• Conditions evaluate correctly. Conditions such as input.flag == true now take the right path, and mistyped names produce a clear error.
• Excel exports keep inputs as text. A workflow input that starts with = is exported to Excel exactly as typed, instead of being treated as a formula.
• Templates Store shows the right projects. When you install from the Templates Store, the project picker lists only projects in your current organization.