Skip to content
Rival Docs

Chatting with an Agent

Overview

The fastest way to test and use a Rival Agent is to chat with it. There are two ways in:

  • Testing - from My Tools, the owner chats with their own agent to test it end-to-end, both before and after publishing.
  • Trial - from the Marketplace, anyone can try a published agent through its public trial chat.

The chat surface supports text, voice, file attachments, interactive outputs (artifacts), and memory across the conversation.

Testing from My Tools

When you open your own agent from My Tools, you get a testing chat. It lets you talk to your agent in its current state - same persona, same guardrails, same tools and connectors you’ve configured. Use it to:

  • Sanity-check the persona and guardrails before you publish.
  • Walk through a real task end-to-end to see whether the right tools fire.
  • Spot a missing memory source or a missing connector.

Each testing session is a normal conversation - you can scroll back, attach files, and use voice the same way you would in a published chat.

Trial from the Marketplace

Once an agent is published, anyone can try it from its listing in the Marketplace through a public trial chat. This is how prospective users get a feel for an agent before adopting it.

The trial is meant for evaluation, so it’s capped - each user gets a limited amount of trial usage per agent. To use agents beyond the trial, you’ll need a plan that includes agent usage; see Plans for what each tier includes.

Sending messages

Type into the composer at the bottom of the chat and press Enter to send. Responses stream in token by token. The agent has memory across the conversation, so you can refer back to earlier turns naturally.

For very long conversations, Rival automatically condenses earlier history server-side so the agent doesn’t lose context - you don’t have to do anything to make this work.

Voice input

You can speak to your agent instead of typing.

  1. Click the microphone icon in the composer.
  2. Speak your message - a pulsing mic and elapsed timer show you’re being recorded.
  3. Click the mic again to stop.
  4. The transcript drops into the input box. Review and edit, then hit Send.

Voice is input only - the agent’s replies are text. If your browser blocks microphone access, a tooltip explains how to enable it.

File attachments

Attach files to give your agent context - paste an error log, share a PDF, drop in a screenshot.

How to attach

  • Click the paperclip icon in the composer.
  • Drag and drop a file onto the input area.
  • Paste an image directly from your clipboard.

Supported file types

CategoryFormats
ImagesJPEG, PNG, GIF, WebP
DocumentsPDF
TextTXT, CSV, MD, JSON, YAML/YML
Code.py, .js, .html, .xml

Unsupported types are rejected with a clear message. Word docs, spreadsheets, ZIPs, and video files are not currently supported.

Working with attachments

  • Attachments appear as chips above the input - remove with the × before sending.
  • You can send a message with only an attachment, no text.
  • Rival pre-processes images and PDFs so the agent can read them regardless of which model is selected.

Artifacts panel

When your agent produces rich output - a chart, a report, a form, a dashboard - it surfaces as an artifact instead of a wall of text.

Two layouts:

  • Inline artifacts appear inside the chat bubble as a bordered card. Forms and quizzes you’re meant to fill in show up this way. Click Open in panel (the maximize icon) to view it larger in the side panel.
  • Panel artifacts open in a right-hand panel alongside the chat. Charts, dashboards, reports, and code previews use this. The chat shows a clickable card; clicking opens the full artifact in the panel.

When you submit an interactive form, the result is populated into the chat input as text. Nothing is sent automatically - review, edit, then hit Send.

Tool calls in the conversation

When your agent calls a tool, you’ll see a tool-call card in the chat showing the tool name, the input arguments (while running), and the output once complete. Status is one of:

  • Success - the tool returned a result.
  • Failed - the tool errored; the message explains why.
  • Insufficient balance - see the out-of-balance flow.

Sub-agent calls

When an agent calls a sub-agent, a card labeled Sub-agent: <name> appears in the chat. The sub-agent’s own internal tool calls are hidden to keep the conversation readable. See Sub-agents for the full model.

Memory

The agent remembers prior turns within a conversation. You can scroll back, reference earlier messages, and continue where you left off. Memory is per-conversation and tied to your account - other users chatting with the same published agent have their own independent conversation history.

Next steps