Agent Studio list page

Location: Settings > Agents

The list page displays all agents in your workspace as a card grid.

Agent card

Each card shows:

• Name: The agent's display name.

• Goal: The agent's high-level objective (truncated).

• Version: Current version label (for example, "Version: Draft" or "Version: 1.0").

Header actions

• Create New Agent: Creates a new agent with default configuration.

• Settings (gear icon): Opens a menu with access to the Datasets page.

Pagination

The list uses cursor-based pagination. Navigation controls appear at the bottom when there are more agents than fit on one page.

Build tab

Location: Settings > Agents > [Agent Name] > Build

The Build tab is where you configure all aspects of your agent.

Agent identity

• Agent Name: Display name shown to users. Click to edit; saved on blur.

• Agent Goal: High-level objective. Auto-resizing text area, saved on blur. Reverts if left empty.

Capabilities section

The Capabilities section contains three configurable areas: Knowledge, Skills, and Guardrails.

Knowledge

Defines which DevRev object types the agent can search.

Available object types:

Each knowledge source appears as a chip showing the object type's icon and name. Click × to remove.

Skills

Defines the actions the agent can perform.

Skill types:

Operation configuration fields:

Guardrails

Defines rules and boundaries the agent must follow.

Guardrail fields:

Guardrail states:

• Enabled: Guardrail is actively enforced.

• Disabled: Guardrail exists but is not enforced.

• Always on: Cannot be toggled off (Default Guardrail only).

Default Guardrail: Every agent has a built-in Default Guardrail that is always active. It provides baseline safety behavior and cannot be disabled or deleted.

Instructions section

A rich text editor for writing detailed behavioral guidance.

• Rich text formatting: Supports Markdown-style formatting.

• @ mentions: Reference specific knowledge sources, tools, or skills inline.

• Auto-save: Changes are saved automatically as you type.

Top-line controls

Controls that appear in the Build tab header:

• Version status badge: Shows the current state—Draft or Live.

• Version History toggle: Opens the version history side panel.

• Playground toggle: Opens the chat playground side panel.

• Publish button: Publishes the current draft version.

• More options menu: Agent-level actions (Duplicate, Delete).

Test tab

Location: Settings > Agents > [Agent Name] > Test

The Test tab provides tools for validating agent behavior.

Preview tests sub-tab

Interactive testing through the Playground.

• Start New Chat: Opens a fresh Playground conversation.

• Test session list: Shows previous test conversations.

• View Trace: Opens the execution trace for a session.

Clicking a test session row opens the Playground loaded with that session's conversation.

Bulk tests sub-tab

Structured testing against datasets.

• Create Bulk Test: Opens the bulk test creation form.

• Bulk test list: Shows all bulk tests with status and results.

Create bulk test form

Evaluators

• Correctness: Measures whether the agent's response accurately addresses the input. Returns a score from 0.0 to 1.0 with an explanation.

• Completeness: Measures whether the response fully covers the expected output. Returns a score from 0.0 to 1.0 with an explanation.

Bulk test results columns

Bulk test statuses

• Queued: Test entry is waiting to be processed.

• Running: Agent is processing this test entry.

• Completed: Test entry finished successfully.

• Errored: Test entry failed to complete.

Observe tab

Location: Agent Studio > [Agent Name] > Observe

The Observe tab provides monitoring and analysis tools for agent performance and conversation history.

Analytics sub-tab

The Analytics sub-tab displays a performance dashboard for the agent. The dashboard has three possible states:

• Not enabled: Shows "Metrics collection for this agent is not enabled" with an Enable Evaluation button.

• Enabling: Shows a loading state while analytics is being set up.

• Active: Displays the analytics dashboard filtered to this agent, defaulting to the last 7 days.

Sessions sub-tab

The Sessions sub-tab shows all conversations the agent has participated in.

Session list columns:

• Trigger: What initiated the conversation (not sortable).

• Members: Participants in the conversation (sortable).

• Last message: Timestamp of the most recent message (sortable).

Session list filters:

• Last message: Filter by time range.

• Trigger: Filter by conversation trigger type.

Session detail view

Clicking a session opens its detail page, which contains the following sections:

• Session header: Participants, timestamps, and trigger information.

• Conversation: Full message history.

• Execution trace: Step-by-step agent reasoning for each message.

Trace elements

Each execution trace can contain the following elements:

• Thought: The agent's internal reasoning process.

• LLM Reasoning (intermediate): Reasoning steps during processing.

• LLM Reasoning (final): Final reasoning before responding.

• Input: Data sent to a skill or knowledge source.

• Output: Data received back from a skill or knowledge source.

• Guardrail Check: Guardrail evaluation results.

• Response Time: Duration of each processing step.

• View Workflow: Link to view the executed workflow, if applicable.

Version management

Version states

• Draft: Work-in-progress version that is freely editable. Only one draft exists at a time.

• Live (Published): The active version serving users. Only one live version exists at a time.

• Archived: A previously published version, preserved for history.

Version actions

• Edit: Modify any field on a draft version. Editing a live version auto-creates a new draft.

• Publish: Promote the current draft to live. This archives the previous live version.

• Restore: Create a new draft from any previous version's configuration.

Version history panel

Access the version history panel via the version history toggle in the Build tab header. The panel contains:

• Version timeline: Chronological list of all versions.

• Version label: Displayed in the format "V{number} · {status}".

• Status events: Created, Edited, Published, and Archived — each with the user and timestamp.

• View Older History: Loads more versions beyond the initial set.

• Actions menu: Per-version actions such as Restore.

Publish confirmation dialog

When publishing, a confirmation dialog appears with the following elements:

• Version number: The version being published.

• Current draft label: Identifies the draft being promoted.

• Publish button: Confirms and publishes the version.

• Cancel button: Returns without publishing.

Restore confirmation dialog

When restoring, a confirmation dialog appears with the following elements:

• Target version: The version being restored, including its number and state.

• Description: Explains that a new draft will be created from this version.

• Restore button: Confirms and creates the new draft.

Datasets

Location: Agent Studio > Settings (gear) > Datasets

Datasets contain collections of test cases used for bulk testing of agents.

Datasets list

• Name: Dataset name.

• Description: Optional description.

• Entries: Number of test cases.

• Test Runs: Number of bulk tests run against this dataset.

• Created By: User who created the dataset.

• Date Created: Creation timestamp.

Import dataset dialog

To import a dataset, provide the following fields:

• File (required): CSV file upload via drag-and-drop or browse. The file must not exceed the maximum size limit. The CSV must contain three columns: Input, Expected Output, and Remarks. A downloadable CSV template is provided in the dialog to show the expected format.

• Name (required): Display name for the dataset.

• Description (optional): Context about what the dataset tests.

Dataset detail page

The detail page contains two tabs:

• Entries: List of test cases showing Input, Expected Output, and Remarks columns.

• Completed Tests: Bulk tests that have been run against this dataset.

Available actions on the detail page:

• Run Test: Opens the Create Bulk Test form with this dataset pre-selected.

• Owned By: Shows the dataset owner.

Agent types

Agent type is selected during creation and determines the agent's access level and available channels.