Your Assistant
Your assistant is the heart of your Yedric setup. It holds everything that defines how the assistant behaves for your users. This is what it knows, how it responds, what it can do, and how it looks inside your app.
Each tab in your assistant controls a different part of that experience.
Dashboard
An overview of how your assistant is performing. When you first create an assistant, this shows a setup checklist. Once conversations are coming in, it displays conversation volume, sentiment trends, common topics, and AI-generated suggestions for improving your instructions and filling gaps in your knowledge base. You can apply instruction suggestions directly from here without switching tabs.
Start here when you want a quick read on whether your assistant is working well, or when you want the AI to surface things worth fixing. See Dashboard for a breakdown of every metric and recommendation.
Prompt
The system prompt that shapes how Yedric thinks and responds. Every conversation starts with this. It tells the assistant what it is, what it can help with, how it should talk to users, and what it should avoid. Changes here affect every future conversation.
This is the first thing to set up, and often the most impactful thing to revisit when the assistant is not behaving the way you want. See Prompt for guidance on writing an effective system prompt.
Knowledge Base
The content Yedric draws on to answer your users' questions. You can add help articles, documentation pages, PDFs, or plain text. The assistant searches this content automatically when it needs to answer a question, so you do not need to configure anything for that to work.
Add content here whenever users are asking questions the assistant should be able to answer but is not. See Knowledge Base to learn about the different ways to add and manage content.
Actions
The actions Yedric can take beyond just responding. Built-in tools include searching your knowledge base, asking the user a clarifying question, displaying a table in chat, and handing off to a human agent. If your developer has connected a custom MCP server, additional tools from your own app or services will appear here too.
For each tool you can control whether it runs automatically, requires the user's confirmation, or is disabled. See Actions for a full breakdown of what each built-in tool does and how to add custom ones.
Page Prompts
Page prompts that let Yedric adapt based on where a user is in your app. You can show different suggestion buttons on different pages, and optionally override parts of the system prompt for specific URLs. A user on your pricing page might see different suggestions than one on your settings page.
Come here once your assistant is working well and you want to make it feel more tailored to different parts of your product. See Page Prompts to learn how to write URL patterns and add per-page suggestions.
Conversations
A log of every interaction users have had with your assistant. You can read individual conversations, tag them, vote them up or down, and use filters to find patterns. The Trends view gives you an aggregated view of what users are asking and how they are feeling about the responses.
This is where you go to understand what is actually happening between your users and the assistant. See Conversations for more on tagging, filtering, and using Trends.
Alerts
Proactive messages the assistant shows on your site without waiting for the user to open the chat. Use them to announce a feature, run a promotion, or nudge users toward the assistant on specific pages. You can target alerts by page or by user, cap how often they show, and add a button that opens a link or drops the user straight into a conversation.
Come here when you want the assistant to reach out first rather than only responding. See Alerts for the placement types and full list of options.
Widget
Controls how the assistant looks and behaves inside your app. You can set the theme, colors, font, greeting message, suggestion style, position on screen, and whether the widget opens automatically. Changes preview live in the panel so you can see them before saving.
See Widget for the full list of customization options, and the Developers section for the complete attribute reference.
Simulations
A testing sandbox for trying out changes before they go live. You can run a saved simulation against your current assistant to check whether it still behaves as expected, or run an ad-hoc conversation to test something specific. Simulations can be organized into suites and run on a schedule for ongoing regression testing.
Use this before publishing significant prompt changes or after updating your knowledge base. See Simulations for how to set up test scenarios and success criteria.
Versions
A history of every saved change to your assistant's configuration. Each time you save changes, a new version is recorded. You can view what changed between any two versions, restore a previous version, and sync your configuration to a Git repository.
Useful if you need to roll back a change that made things worse, or review what was different before a drop in quality. See Versions for more on diffing and restoring versions.
Settings
Assistant-level configuration: the assistant's name, the Conversation tagging controls that decide how the AI analyzer tags its conversations (auto-tagging frustrated conversations, frustration sensitivity, feature-request detection, and your own analyzer instructions), the Live chat & support setup, and assistant-level secrets. Secrets are environment-variable-style values you can reference in your instructions or tools, useful for passing credentials without hardcoding them. Provider API keys are not here, they are set once per organization. See Settings.
If you have not set up your assistant yet, start with Getting Started.
Last updated