Agents

Agents are a combination of user specified instructions, knowledge, and Actions. Think of them as AI team members that are optimized for a specific task. Onyx Agents can be for individual use or shared with other users or user-groups within Onyx.

When to use an Agent

Use agents to build the best experience for repeat use cases, especially those that can be shared with other members of the team. Some common examples include:

Onboarding Assistant
AI Helpdesk
Engineer Copilot
Legal Reviewer

These are typically scoped down to a particular task such as reviewing documents to ensure they comply with security policies, filling out RFPs based on previously completed ones, etc.

Creating an Agent

Instructions (Prompting)

The Instruction field is infinitely flexible so it is recommended to test different variations and verify the behavior before sharing the Agent broadly with your team.

Provided below for inspiration is the Instruction for the default Onyx experience.

Onyx system default Instruction

You are a highly capable, thoughtful, and precise assistant. Your goal is to deeply understand the user's intent, ask clarifying questions when needed, think step-by-step through complex problems, provide clear and accurate answers, and proactively anticipate helpful follow-up information. Always prioritize being truthful, nuanced, insightful, and efficient.
The current date is [[CURRENT_DATETIME]]

You use different text styles, bolding, emojis (sparingly), block quotes, and other formatting to make your responses more readable and engaging.
You use proper Markdown and LaTeX to format your responses for math, scientific, and chemical formulas, symbols, etc.: '$$\n[expression]\n$$' for standalone cases and '\( [expression] \)' when inline. For code you prefer to use Markdown and specify the language.
You can use Markdown horizontal rules (---) to separate sections of your responses.
You can use Markdown tables to format your responses for data, lists, and other structured information.

Note that [[CURRENT_DATETIME]] is dynamically replaced and there may be additional Action use instructions depending on the enabled Actions.

One of the best parts of creating an Agent is the ability to control the goal/behavior of the Agent. The instructions are purely in natural language and can include things like:

Always provide your results in a table.
Try to quote directly from the documents rather than rewording anything.
Flag information from documents that are older than 3 months.

Knowledge

It is typically best to include only the necessary knowledge for the Agent to complete its task. The more scoped down the documents, the more reliable the performance.

When creating an Agent, you can include both knowledge from connectors or file upload. The knowledge from connectors will be kept up to date as documents change in the external sources.

Actions

If any of the built-in actions (Web Search, Internal Search, Image Generation, Code Interpreter) are missing, contact your Onyx admin. These must be set up in the admin-panel before being made available in the Agent creation flow.

Actions (sometimes known as Tools) allow the Agents to do things in external applications via APIs. Some tools help Onyx give better answers (such as the Onyx built-in tools) while others allow the system to do more than answer questions. Some example common Actions include:

Updating ticket statuses based on the conversation and if the user’s needs were addressed.
Checking the status of a service to give real-time answers.
Moving deals along a sales pipeline in a CRM based on call transcripts or user request.

Getting Started

Core Features

Onyx Anywhere

Miscellaneous

When to use an Agent

Creating an Agent

Instructions (Prompting)

Knowledge

Actions

Getting Started

Core Features

Onyx Anywhere

Miscellaneous

​When to use an Agent

​Creating an Agent

​Instructions (Prompting)

​Knowledge

​Actions

When to use an Agent

Creating an Agent

Instructions (Prompting)

Knowledge

Actions