Skip to main content
Miraxa is the AI layer that runs alongside everything you build in Spojit. Open it from the navigation bar on any page. It has live context on your workspace, so it can tell you what is running right now, build and edit workflows on your canvas, trace a failed run back to its cause, and point out steps that are costing more than they need to. Miraxa is available on every plan, and its work draws on your monthly AI credit allowance like any other AI step.

​
What Miraxa does

Builds from a sentence

Describe an outcome and Miraxa builds the workflow: trigger, branches, actions and code. It can read vendor API docs to wire up a tool with no pre-built connector.

Knows what is running

Live context on your workflows, runs and connections, so it can answer what is happening now and what changed.

Investigates failures

Point it at a failed execution and it traces the cause across steps, then repairs the workflow or tells you exactly what broke.

Explains a workflow

Ask what an inherited workflow is for and it explains in plain language instead of leaving you to read the graph.

Debugs and improves

Edits live workflows in place, repairs broken steps, and hardens them with retries and error handling.

Finds optimizations

Spots slow, costly or redundant steps and suggests fixes, such as moving a step to Direct Mode or running paths in parallel.

​
Modes

Miraxa runs in one of four modes, chosen from the selector in the chat header. The mode controls how much it is allowed to do on its own, and it is remembered per conversation.
Plan mode is genuinely read-only: mutating tools are withheld from Miraxa rather than merely discouraged, so it cannot change anything even if you ask it to. It will tell you to switch modes instead.
Operator mode is only visible if you hold the permission for it. See Operator Mode for what it can reach and how approvals work.

​
Context awareness

Miraxa knows which page you are on and adapts to it: In the workflow editor it can see everything on the canvas, including node types, labels, how they are wired and each node’s configuration. That is what makes questions like “what does the third step do?” or “is there a problem with my loop?” answerable.

​
Building and editing workflows

In the workflow editor, Miraxa changes the canvas directly. You can ask it to:
  • Add nodes: “Add a Slack connector after the trigger”
  • Remove nodes: “Delete the transform node”
  • Update configuration: “Change the condition to check if order total is over $100”
  • Connect nodes: “Connect the Shopify node to the condition”
  • Build whole flows: “Fetch new Shopify orders and send a Slack notification for each one”
  • Harden a flow: “Add retries and an error path to the payment step”
Nodes appear, move and connect in real time, and Miraxa explains what it changed. After a build or edit it validates the workflow automatically, so you find out immediately if something does not compile.
If you are not sure where to start, describe what you want in plain language and let Miraxa scaffold the whole thing. Refining a rough workflow is faster than starting from an empty canvas.

​
Attachments

Miraxa can see what you send it. Paste an image straight from the clipboard, drag a file anywhere onto the chat panel, or use the paperclip in the composer. It works in every mode and on every page. The most useful thing to do with it is skip the describing. Screenshot a workflow from another tool, photograph a process someone sketched on a whiteboard, or paste a diagram, and ask Miraxa to build it. It reads the picture and starts building on the canvas, which is usually faster than writing out every step in prose. Attachments are available to anyone who can use Miraxa, with no separate permission.
For a large or heavily tabular file, upload it to your File Store instead and ask Miraxa to read it from there. That keeps the conversation small and gives you the file again later.
Anything you attach can be kept: ask Miraxa to save it to your File Store and it becomes an ordinary file in your workspace rather than something that disappears with the conversation.

​
Other things to ask

  • Explanations: “What does this workflow do?” or “Explain the condition node”
  • Troubleshooting: “Why did last night’s run fail?” or “What is wrong with my loop?”
  • Cost and speed: “Which steps here are burning the most credits?”
  • Guidance: “How do I connect to Shopify?” or “What is the best way to loop over an array?”
  • Discovery: “What connectors are available?” or “How do schedules work?”

​
Choosing a model

The chat header has a model picker, and your choice persists across page reloads. The list is deliberately short: only the most capable models are offered here, because building a correct workflow in one pass is a harder job than answering a question, and the faster models are not reliable enough at it. Those faster models remain available inside workflow steps, where you choose per node. Every plan defaults to Claude Opus 5 for Miraxa. See AI Models for what each model is good at, and Governance for how an administrator can change the default or restrict the list for a workspace.

​
Getting good results

  • Be specific: “Add a Slack connector that posts to #general” beats “add a messaging step”
  • Name your nodes: if you have labelled them, refer to those labels
  • Iterate: start rough and refine step by step rather than specifying everything up front
  • Ask it to explain: if you are unsure what it did, ask before you run anything
  • Match the mode to the risk: Plan to look, Ask to proceed carefully, Auto to move fast
  • Use it anywhere: Miraxa is not only for the editor. Ask about connections, runs or settings from any page