Super Agent

Super Agent is MuleRun's main entry point and is built around "automating knowledge work." Unlike a traditional Q&A-style AI assistant, it can call browsers, editors, image generation, databases, and other tools to take a task from breakdown all the way to delivery. Chat is its core, default page.

Open it at: https://mulerun.com/chat

What Is It Good For?

• Information organization, research, summaries, and reports.
• Copywriting, emails, proposals, slide decks, PDF summaries.
• Data analysis, stock and crypto research, e-commerce and competitor research.
• Generating images, videos, websites, landing pages, mini-games, portfolios.
• Long-running automation workflows when paired with MuleRun Computer.

Watching the Agent Work

Every Agent reply is built from expandable atoms, so you can see what it is thinking, doing, and producing in one stream:

• Thinking: the Agent's reasoning before it acts.
• Plan: the steps it intends to take.
• Tool Call (and Tool Call Group): each tool invocation, with inputs and results; grouped when several run together.
• AskUserQuestion: a question you must answer before the Agent can continue.
• Authorization Card: a prompt to approve an action that needs your permission.
• File Result: files the Agent generated or modified.

The Input Box

• Attachments: click + to attach a local file or pull one From Drive. Multimodal input is supported (image, document, audio, video, code, zip).
• Slash commands: type / to open the command palette.
• Office Sync: when you upload an Excel or PowerPoint file, toggle Office Sync so the Agent's edits write back to the source file bidirectionally.
• Schedule Chip: click the calendar icon before sending to register the message as a Scheduled Task that runs on a cron or natural-language schedule.

Single-Agent vs Parallel Mode

By default a message runs on a single Agent. Switch to parallel mode to run the same message across multiple Agents/models at once and compare their results side by side. Parallel mode is paid-only.

Installing a Skill in a Message

You can install and use a Skill inline in one message:

/skill-finder install skill_id directly @{owner}/{repo}#{skillName}:{version} {your actual message}

The command prefix is hidden in the sent bubble, leaving only your actual message visible. You can stack multiple Skills in the same message.

Tasks and Sessions

One full conversation is one Task (also called a Session). You state a goal, the Agent works through it end-to-end, and the whole conversation is the Task. Titles are generated automatically.

The sidebar keeps your history with search. Use the "…" menu on any session to rename, favorite, share, or delete it.

Stop and Message Actions

• Stop: interrupt a running Agent at any time; the output produced so far is kept.
• Hover any message to copy, regenerate, branch, or give a thumbs up / down.

Share and Save as Template

• Share: generates a read-only link (with an optional access code). Recipients can view the full history but cannot continue the chat — real-time collaboration requires a Team workspace.
• Save as Template: stores the session as a template you can reuse with one click from New Chat.

Deleted Sessions

Deleted sessions are recoverable within 30 days, after which they are permanently gone.

How Are Credits Charged?

Credits are charged based on the model used plus the tools invoked (search, code execution, file operations, and so on). Each Tool Call card shows its estimated and actual consumption. Capabilities and quotas vary by plan — see the pricing page for details.

How Can I Reduce Credits Usage?

• Avoid endlessly extending one very long conversation.
• At the end of each stage, ask the Agent to summarize 3-5 key points.
• When starting a new conversation, include only the necessary context.
• Ask for one clear goal at a time.
• Avoid unnecessary long context or repeated files.

How Do I Rate the Agent's Answers?

Every Agent response has a star-rating control. When you give a low rating, you can pick a specific dimension: UI issue, Agent stuck, over-refusal, didn't follow instructions, factual error, incomplete answer, too expensive, or inappropriate content, and add a comment. The feedback goes directly into MuleRun's improvement loop.