Building with RivalBot

What is RivalBot

RivalBot is the AI assistant inside Studio (/studio). Instead of starting from a blank editor, you describe what you want in natural language and RivalBot generates a working tool - code, input schema, test cases, metadata, and icon - then hands it to you in Studio for review and publish.

You’ll also see RivalBot inside Studio itself, where you can chat to edit an existing tool.

Building a tool from Studio

Sign in at cortexone.rival.io and open Studio from the sidebar.
In the “What do you want to build?” prompt, describe your tool.
Answer any clarifying question RivalBot asks.
Wait for it to finish generating - you’ll land in Studio with everything pre-filled.

Example prompts:

“A tool that takes a URL and returns the page’s title and meta description.”
“Summarize a CSV by computing column-level stats and returning the top 5 outliers.”
“Tag incoming customer emails by sentiment and urgency.”

RivalBot streams status messages while it works - “Thinking…”, “Generating code…”, “Generating documentation, metadata, events & icon…”, “Creating tool…” - and asks a clarifying question if your prompt is ambiguous instead of guessing.

When it’s done, Studio opens with code, test cases, metadata, and documentation already drafted. Review and adjust across the Studio tabs, then publish.

Attachments, connectors, and voice

Beyond typing, the RivalBot chat box gives you three extra inputs:

Attach files - add reference files with the attach button, or paste them straight into the box. Supported types include images (.png, .jpg, .gif, .webp), documents (.pdf, .txt, .csv, .md), and structured or code files (.json, .yaml, .py, .js, .html, .xml). RivalBot uses them as context while it builds.
Connectors - open the Connectors panel to link third-party services (via OAuth) that your tool should be able to call.
Voice input - tap the microphone button to dictate your prompt instead of typing; tap again to stop.

Editing an existing tool

Open any tool you own and you’ll see a RivalBot chat panel inside Studio. Use it to iterate without leaving the page.

Common quick actions:

“Improve the code”
“Add error handling”
“Write tests”
“Explain logic”
“Debug this error”
“Generate documentation”
“Generate a changelog”

RivalBot can update any part of the tool - code, test cases, metadata, documentation, icon - based on what you ask. Conversation history is preserved per tool, so you can come back later and pick up where you left off.

RivalBot finds existing tools

Before generating new code, RivalBot searches the Tools Marketplace for tools that already do what you’re asking. Two behaviors result:

Build pre-check

When you ask RivalBot to build something that already exists, it surfaces matching marketplace tools first as cards. You choose:

Use existing - add the existing tool to your org
Build anyway - keep going with the build

This saves time and avoids reinventing tools the community has already published.

Inline tool execution in chat

When you’re chatting with RivalBot about something that doesn’t need a build at all - for example “Give me the latest AI news” - RivalBot may call a marketplace tool directly inside the conversation and stream the result back. You’ll see a tool-call card showing:

Tool name
Input arguments while running
Output when complete

Status is one of Success, Failed, or Insufficient balance (the last one routes you to top up your Run Balance).

RivalBot may also show a Related tools chip row below its replies - marketplace tools that matched but weren’t called. Click one to open its detail page.

Status messages

Some of the status lines you’ll see and what they mean:

Status	Meaning
”Thinking…”	RivalBot is figuring out what to do.
”Generating code…”	Writing the handler code.
”Improving code…”	Refactoring or fixing existing code.
”Debugging code…”	Investigating an error you reported.
”Generating test cases…”	Creating example events you can run against your tool.
”Updating metadata…”	Editing the tool’s category, sectors, tags, etc.
”Generating documentation…”	Writing or rewriting the tool’s docs page.
”Generating changelog…”	Drafting release notes from your changes.
”Checking marketplace…”	Looking for existing tools that already solve this.
”Analyzing requirements…”	Breaking your prompt down before building.

Tips for working with RivalBot

Start specific. “A tool that takes X and returns Y” is faster than “Something to help with marketing.”
Iterate in small chunks. Ask for one change at a time so you can see what RivalBot did and easily revert.
Test as you go. Every change is also reflected in test cases - run them from the Preview tab to make sure the tool still works.
Don’t worry about the first draft. RivalBot’s output is a starting point. Edit by hand or by asking RivalBot to update it.