Prerequisites:

• Access to DevRev with permissions to create agents.

• At least one knowledge source, for example, articles or tickets in your DevRev workspace.

• If you plan to attach workflow skills, ensure you have already built the workflows you intend to use.

Step 1: Open Agent Studio

In the DevRev sidebar, go to Settings > Agent Studio. The Agent Studio page displays a grid of agent cards, or an empty state if no agents exist yet.

Step 2: Create a new agent

Click Create New Agent in the top-right corner of the Agent Studio page. You are taken to the Build tab for your new agent.

Your agent is created with a default name ("New Agent"), a default goal, and a default guardrail already enabled.

Step 3: Name your agent and set a goal

At the top of the Build page, you see two editable fields:

1. Agent name: Click the name at the top and type a meaningful name for your agent, for example, Support Assistant.

2. Agent goal: Below the name, click the text area and describe what your agent should do. For example:

   You are a customer support assistant. Help users resolve their issues by searching the knowledge base and creating tickets when necessary.

The goal gives the agent its overall direction. Be specific about the agent's purpose.

Step 4: Add knowledge sources

Knowledge sources tell the agent where to look when answering questions.

1. In the Capabilities section, find the Knowledge row and click + Add.

2. In the Add Knowledge modal, select the object types you want the agent to search. Common choices include:

   • Article: Knowledge base articles.

   • Ticket: Support tickets.

   • Question & Answer: Q&A entries.

   • Conversation: Past conversations.

3. Use the search bar to filter the list if needed, then click Add to confirm your selection.

Your selected knowledge sources appear as chips in the Knowledge row.

Step 5: Attach skills

Skills give your agent the ability to take actions such as creating tickets, sending messages, or running workflows.

1. In the Capabilities section, find the Skills row and click + Add.

2. In the Add Skill modal, browse or search using the tabs:

   • All: A combined view showing operations and workflows.

   • Operations: Built-in DevRev actions (e.g., create a ticket, update an issue).

   • Workflows: Custom automation workflows you have already built.

3. For operations: After selecting one, a configuration panel opens where you can:

   • Give the skill a name (letters, numbers, and underscores only).

   • Add a description of when the agent should use this skill.

   • Configure input fields: toggle Auto-fill to let the agent determine values from context, or set manual defaults.

   • Toggle Execute as User to run the operation on behalf of the end user rather than the agent.

   • Configure Connections if the operation requires external integrations or keyrings.

4. For workflows: Select the workflow to add it directly.

5. Confirm to add the skill.

Your attached skills appear as chips in the Skills row.

💡 Tip: If you do not see any workflows in the Workflows tab, you need to create them first before you can attach them as skills.

Step 6: Review guardrails

Guardrails define boundaries for what the agent can and cannot do. A Default Guardrail is already enabled for every new agent. The Default Guardrail is always on and cannot be toggled off or deleted.

1. In the Capabilities section, find the Guardrails row. You see the Default Guardrail with an Always on label.

2. To add a custom guardrail, click + Add:

   • Enter a topic name: a short label for the guardrail.

   • Select a type: the category of restriction.

   • Write a description: the rule the agent must follow (e.g., "Never share internal pricing information with customers").

3. Click Create to add it.

You can toggle custom guardrails on and off, edit them, or delete them at any time.

Step 7: Write instructions

Instructions provide detailed guidance for how the agent should behave. Think of them as the agent's playbook.

1. Scroll down to the Instructions section.

2. Click the editor area and write your instructions using rich text. Use @ to reference specific tools or skills directly in your instructions.

Tips for good instructions:

• Be specific about tone and style ("Always be polite and professional").

• Define escalation rules ("If the user asks about billing, create a ticket for the billing team").

• Specify what the agent should not do ("Do not make promises about delivery timelines").

Step 8: Test your agent in the Playground

1. Click the Test tab at the top of the page. You see the Preview Tests sub-tab with a prompt to start testing.

2. Click Start New Chat. A chat panel opens on the right side of the screen.

3. Type a message to your agent and press Enter. Watch as the agent responds using the knowledge and skills you configured.

4. To inspect the agent's reasoning, click View Trace on any response. The execution trace shows which skills, knowledge sources, and guardrails the agent invoked and the decisions it made at each step.

Try a few different questions to see how your agent handles them. If the responses are not what you expect, go back to the Build tab and refine the goal, instructions, or skills.

Step 9: Publish your agent

Once you are satisfied with your agent's behavior:

1. Go back to the Build tab and click Publish in the top-right corner.

2. In the confirmation dialog, review the version number that will be published, then click Publish to confirm.

Your agent is now live. The version status changes from Draft to Live. Any previously live version is automatically archived. You can view and restore archived versions from the version history.

Next steps

Now that you have built and published your first agent, explore the following topics in the Agent Studio task reference:

• Run bulk tests to evaluate your agent against a dataset of test cases.

• Monitor performance in the Observe tab to track analytics and review session traces.

• Iterate with versions to make improvements without disrupting the live agent.

For a complete field-level reference of every feature, see the Agent Studio reference.