Purpose
This article is designed to guide you through the initial setup of Personal Agents within Purple, by providing a step-by-step overview for getting started. This article explains the concept of Personal Agents, highlights their role in automating tasks and workflows, and details the configuration process, including how to create, customize, and manage these agents according to individual preferences.
On this page:
- Identify the Need for an Agent
- Log Into Purple
- Create and Configure the Agent
- Use and Manage Your Agent
Identify the Need for an Agent
Deciding to launch a new agent is a strategic move. It should be driven by a clear need: whether automating support, surfacing knowledge, or a specific business goal.
Evaluation Checklist:
- What problem will the Agent solve?
- Who is the intended audience?
- What Knowledge Sources are required?
Login to Purple
- Go to: https://purple.uw.edu
- Click the "Sign in with Microsoft" button.
- Enter your UW email address and password when prompted.
- Complete authentication with Duo.
- New to Duo? View Duo guides
Create and Configure the Agent
Step 1: Create the Agent
- In the left navigation pane, click Explore Agents. Next to My Agents, click + Create. A Quick Create popup window displays.
- In the Agent Name field, enter a name that can be easily referenced later on, and click Let's go. The Create Personal Agent page displays. This provides an all-inclusive view of your Agent that details configuration options, branding, and more.
Step 2: Add Agent Instructions
Agent Instructions are used to define boundaries, policies, and user expectations for the Agent, giving compliance, security, and improvements to the user experience as well. Default instructions are provided but can be edited to suit your Agent's purpose.
Step 3: Add Knowledge Sources
Knowledge Sources provide the Agent with additional information that can be sourced from multiple locations. You can:
- Upload relevant documents or files. Your Agent will have access to information located within these files.
- Direct the Agent to a website. Use this option when the Agent needs information from a specific site or domain. Configure settings such as Maximum Pages to Use and When to Use so the Agent knows when and how to rely on that source. If different subdomains serve different purposes, add them separately with clear usage instructions.
- Limit the number of pages the Agent can use. Set a page limit to give the Agent enough context without overwhelming it. Too many pages can reduce response quality or slow performance. As a best practice, limit each source to about 10 pages when possible.
Step 4: Select Capabilities
Capabilities give the Agent more features within Purple which allows for more usability for created Agents, such as:
- Enable Internet Search
- Allow Add Files & Images (using attachments with the Agent)
- Allow Image Creation
Step 5: Configure the Model
Model Configuration for Assistants Model Type:
The Model will default to what the organization has configured. If you select Change, you can configure:
Maximum Response Length (in tokens) - controls the total allowable answer length. Please ensure the number entered does not exceed the model's built-in maximum response length. The default maximum response length that works for all supported models is 12800 tokens. If you enter a number larger than the model's limit, you may encounter issues receiving a response when the model exceeds that limit.
Creativity: how much randomness and novelty the model introduces.
| Creativity Level | Description |
|---|---|
| None | Fully deterministic. The same prompt produces the same response every time. Focused strictly on factual or canonical answers. |
| Low | Minimal variation. Responses are conservative, predictable, and grounded in known patterns. |
| Medium | Balanced creativity. Allows light variation in phrasing and examples while remaining accurate and controlled. |
| High | Highly expressive and diverse. Responses may include novel ideas, metaphors, or unconventional phrasing, with increased risk of speculation. |
Advanced Settings
If you open up Advanced Settings, you can also modify Top P and Frequency Penalty.
| Setting | Description |
|---|---|
| Top P | Limits token selection to the smallest set of tokens whose cumulative probability ≥ p.
|
| Frequency Penalty | Reduces the likelihood of repeating the same words or phrases
|
Model Configuration for OpenAI Model Type:
| Verbosity Level | Description |
|---|---|
| Auto | Adaptive detail level. Adjusts verbosity based on query complexity: brief for simple questions, detailed for complex topics. |
| Low | Brief and direct. Includes necessary context but avoids unnecessary explanation. Suitable for quick answers and experienced users. |
| Medium | Balanced detail. Provides clear explanations with relevant context and examples where helpful. Default conversational style. |
| High | Comprehensive and thorough. Includes detailed explanations, multiple examples, context, and reasoning. Suitable for learning or complex topics. |
| Reasoning Effort Level | Description |
|---|---|
| Auto | Adaptive reasoning depth. Automatically scales computational effort based on problem complexity—quick for simple queries, deep for complex problems. |
| Low | Basic reasoning. Considers the direct question and provides straightforward answers with minimal checking or alternatives. |
| Medium | Moderate reasoning. Evaluates multiple angles, checks for consistency, and considers common edge cases before responding. |
| High | Deep reasoning. Extensively analyzes the problem, considers alternatives, verifies logic, examines edge cases, and may reconsider initial conclusions. Slower but more thorough. |
Step 6: Preview and Publish the Agent
Put the final touches on the Agent, such as:
- Short Description: Below the name of the Agent, a description can be provided to quickly explain the usage and capability of the given Agent.
- Preview Your Agent: On the right side of the screen, a sandbox is available for you to test and make adjustments in real-time as you create your Agent.
- Welcome Prompt Editor: When a user opens the Agent, the Welcome Prompt is displayed; this is a fully customizable prompt that can be modified using the Edit button on the bottom-right of it.
- Add Chat Starters: Below the welcome prompt, you are able to create a quick prompt for specific scenarios or questions for users of the Agent.
- Click Create to publish your agent.
Use and Manage Your Agent
After your Agent is created, it will appear in the My Agents section. From there, you can manage it by clicking the Edit icon, which displays on the Agent's tile. You can also manage your Agent from the left navigation pane.
To keep your Agent useful over time, review its settings and knowledge sources regularly.
Update Your Agent
Once created, your Personal Agent can be updated or accessed in several ways:
- Edit the Agent from the My Agents section or the left navigation pane.
- Pin the Agent to your sidebar for quick access.
- Share the Agent with others using a shareable link.
Maintain Knowledge Sources
Knowledge Sources should be reviewed periodically to ensure your Agent continues to provide accurate and relevant responses.
- Verify that sources are credible.
- Document and review sources on a regular schedule.
- Remove outdated or irrelevant sources.