Skip to main content

​
What is this

The superun General Agent lets you embed an agent that can chat and get things done inside the app you build: your users ask questions and the agent streams its answers in real time; it can also read documents you upload, remember each user’s preferences, produce Excel / Word / PPT / PDF files, and even perform real business actions such as looking up an order or issuing a refund. No coding is required. Describe what you want in your project chat and superun handles the agent creation, backend wiring, and chat interface automatically; later adjustments are made the same way. Typical scenarios:
  • Knowledge-base Q&A: upload product manuals or policy documents; the agent answers strictly from them and cites the source. If the answer isn’t in the knowledge base, it says so plainly — it never makes things up.
  • Customer service: answers from your company knowledge base and can look up orders, issue refunds, and escalate to a human; every commitment is validated by your system.
  • Office assistant: remembers what each employee has done and reads/produces Excel / PPT / Word / PDF — “turn these numbers into a weekly report”, “what did I work on last week?”.
  • Personal assistant: long-term memory across conversations — preferences and ongoing projects mentioned today are still remembered in a new conversation next week.

​
Step 1: Describe the agent you want

Say it directly in your project chat, for example: 
Build me a Q&A app: users ask questions and the agent streams its answers,
with support for multiple conversations. The agent's role is a "baby-product
shopping advisor" — warm and professional in tone.
The more specific the role, the better the answers: who it is, what it’s good at, what tone to use, and what to do when it doesn’t know.

​
Step 2: superun builds it

superun completes three things automatically:
  1. Creates the agent: role and capabilities set from your description.
  2. Wires up the backend: conversation management and reconnect-on-interrupt are built in.
  3. Builds the chat interface: you can ask it to restyle anything later.

​
Step 3: Verify the result

Send a message to try it out:
  • Replies stream in as they are produced, with a running indicator while the agent is thinking.
  • Refreshing the page, closing the tab, or losing the network all recover automatically — the conversation picks up where it left off with nothing lost.
  • Every conversation is saved independently; come back and continue any time.
After it’s built, adjustments usually come down to three kinds of requests: change the role (“change the agent’s role to…”), change the interface (“make the chat window dark-themed”), and add capabilities (see “Capability add-ons” below).

​
Capability add-ons

The agent ships with a set of built-in tools that work out of the box — nothing to enable:
  • Run code and commands: executes code and processes data in an isolated runtime, e.g. computing a statistics summary on the spot.
  • Read and write files: reads, writes, and edits files in its workspace — cloud files and office document output are built on top of this.
  • Search through documents: finds content fast by filename and text patterns; knowledge-base Q&A relies on it to pinpoint sources.
  • Web search and page fetching: searches the internet and fetches specific pages, e.g. “let it look up the latest information online when answering”.
On top of that, each capability below is opt-in: to enable one, just say so in your project chat.

​
Multiple conversations

Like mainstream AI chat products — a conversation list on the left where users can create, switch, rename, and delete conversations. Just say “add a conversation sidebar to the chat, with support for creating and deleting conversations”.

​
Cloud files

Upload: ground the agent in documents you provide — product manuals, contracts, reports. Say “support uploading PDF / Word documents so the agent answers based on their content”. Download: files the agent produces (reports, documents, etc.) appear in the conversation’s file list, ready to download with one click.
Name uploaded cloud files with English letters and digits only (e.g. product-manual.pdf) — non-English filenames cause the upload to fail. Non-English content inside the file is perfectly fine.
A freshly produced file takes a few seconds to appear in the file list; if the list looks empty, wait a moment and refresh.

​
Office document output

After saying “let the agent generate Excel reports and slide decks”, the agent can produce office files directly — for example “turn this sales data into an Excel report”, “generate a slide deck from this outline”, “export the conclusions of this conversation as a PDF”.

​
Long-term memory

By default each conversation is independent; after saying “give the agent long-term memory so it remembers each user’s preferences and ongoing tasks”, the agent remembers users across conversations — preferences and ongoing tasks mentioned today are still there in a new conversation next week. Memory scope is up to you: private memory per user, a shared memory for a team, or one knowledge pool for all users — just say which when describing your needs.

​
Real business actions

Make the agent not just talk but act: query your database, call external APIs, write business data. For example “the agent should be able to look up order status (from my orders table) and initiate refunds — it must verify the order is eligible before refunding”.
  • Finishes even if the tab is closed: once a user triggers an action, it runs to completion even if they close the page, and the result is never lost.
  • Sensitive actions are validated server-side: refund amounts, eligibility, and similar rules are enforced by your backend; the agent cannot overstep.

​
In-chat confirmation and choices

When a decision is needed, the agent can present buttons, forms, or confirmation dialogs right inside its reply; the conversation continues after the user responds — ideal for refund confirmations, pick-one choices, or collecting details. For example “before refunding, have the agent show a confirmation button and only proceed after the user clicks confirm”.

​
Connecting external systems

The agent can integrate with external services you specify: internal systems, third-party platform APIs, live data sources. For example “let the agent look up ticket status in our internal ticketing system”.

​
Multi-user data isolation

Before opening the app to multiple users, enable user login: each user then sees only their own conversations and files. Just say “enable user login so each user only sees their own conversations”.

​
Things to know

  • English filenames for cloud files: name uploads with English letters and digits; non-English filenames cause the upload to fail.
  • Deliverables appear after a short delay: freshly produced files take a few seconds to show up in the download list.
  • Interruptions take care of themselves: refreshing, losing the network, or closing the page all recover automatically — nothing for you to handle.
  • Conversation data lives in your own project: chat history is stored in your project’s database, fully under your control.
  • Answer quality follows the role and the documents: if answers go off-track, refine the role description first, or upload documents and require “answer only from the documents; say so plainly when the answer isn’t there”.

​
FAQ

No. Describe what you want in your project chat and superun builds everything; later adjustments are made the same way.
Deducted from your superun balance based on actual compute used.
No. When they return, the conversation recovers automatically and the answer continues from where it left off.
The most common cause is a non-English filename. Rename the file in English (e.g. manual.pdf) and upload again.
In that conversation’s file list. Freshly produced files take a few seconds to appear — if the list is empty, wait a moment and refresh.
Upload documents to build a knowledge base and require in the role description that it answers only from the documents and says plainly when it doesn’t know; you can also ask it to cite sources so answers are easy to verify.
Yes. Just say “add another agent whose role is …” — each agent has its own role, and their conversations don’t interfere.