Agents

​

Overview

ELITEA Agents are customizable AI-powered virtual assistants that automate tasks and streamline workflows within the ELITEA platform. Each agent is purpose-built to handle specific tasks or workflows based on custom instructions, integrated toolkits, and AI model capabilities you configure. Agents can interact with external services, make intelligent decisions, and perform complex actions—from creating Jira tickets and managing GitHub repositories to analyzing data and generating documentation. / Why Use Agents? Unlike open-ended AI conversations, agents provide a structured and efficient approach to task automation:

• Task Automation: Automate repetitive and complex multi-step workflows without constant human intervention
• Increased Productivity: Reduce manual effort by delegating routine tasks to intelligent agents
• Consistency: Ensure standardized processes and outputs across your team
• Integration: Seamlessly connect multiple tools and services to create powerful automated workflows
• Scalability: Deploy agents across different domains and scale operations as your needs grow

How Agents Work Creating an agent involves three key components:

1. Instructions: Define the agent’s behavior, goals, and decision-making logic through custom instructions
2. Toolkits & Integrations: Connect external services (GitHub, Jira, Slack, etc.) and internal tools (Attachments, Image creation, Data Analysis, Planner, Python sandbox, Swarm Mode, Smart Tools Selection) to extend capabilities
3. AI Model Configuration: Select and configure the appropriate language model (GPT-4o, GPT-5.1, etc.) with optimal settings for your use case

Once configured, the agent leverages advanced natural language processing to interpret instructions, interact with connected tools, and autonomously execute tasks while adapting to changing conditions. Key Capabilities

• Autonomous Operation: Independently executes tasks and makes decisions based on predefined instructions and goals
• Proactive Problem-Solving: Determines optimal next steps to achieve objectives, even without explicit guidance
• Multi-Service Integration: Combines external toolkits (project management, version control, testing tools) with internal capabilities
• Flexible Customization: Tailor instructions, tools, and model settings to match specific requirements and workflows
• Version Management: Create, manage, and publish different versions of agents for various use cases
• Context Awareness: Maintain conversation context and manage token budgets for efficient, focused interactions

By leveraging ELITEA Agents, you can unlock AI-driven automation that transforms task execution, increases efficiency, and enables your team to focus on strategic and creative work.

​

Navigating the Agents Menu

Click Agents in the main sidebar to open the Agents dashboard — a list of all agents in the current project. The dashboard supports card and table views, search by name or keyword, tag-based filtering, and pinning agents to the top of the list.

---

​

Creating an Agent

To set up a new agent:

1. Click the + Agent button located at the top of the main sidebar.
2. Fill in the agent configuration fields:

   Section	Field	Required	Example
   General	Name	Yes	Code Review Assistant
   General	Description	Yes	Reviews pull requests for bugs, security issues, and style violations
   General	Tags	No	code-review, github, quality
   Instructions	Instructions	Yes	You are a code review expert. Analyze the provided diff for bugs, security vulnerabilities, and style issues. Return findings as a numbered list.
   Welcome Message	Welcome Message	No	Hello! Paste a PR link or diff and I'll review it for you.
   Conversation Starters	Conversation Starters	No	Review this PR for security issues, Check code style compliance
   Advanced	Step Limit	No	50
   Advanced	Ignore Project Context	No	Enable when the agent should focus solely on its own instructions, ignoring project-wide context
   Editor Notes	Notes	No	Internal note: Use this version for PR review experiments only.

3. Click Save.

After clicking Save, the agent configuration page will open. The page uses a two-column layout:

• Left panel — the configuration form with all settings sections (General, Instructions, Variables, Welcome Message, Toolkits, Skills, Conversation Starters, Advanced, Editor Notes, Information)
• Right panel — an embedded live chat panel for interactively testing the agent without leaving the configuration view

From this page you can:

• Add Toolkits, agents, MCPs, and pipelines
• Attach Skills
• Select AI Model and configure model settings
• Enable Internal Tools
• Set Variable Values
• Test Interactively using the embedded chat panel
• View Run History

Your newly created agent will subsequently appear on the Agents page for your project.

​

AI-Assisted Agent Creation (Build with AI)

Instead of filling in the form manually, use Build with AI to generate a complete agent draft from a plain-language description.

1. Open the agent creation form (+ Create → select Agent, or Agents → + Create).
2. In the General section header, click the Build with AI button (✨ sparkle icon).
3. In the dialog, describe what the agent should do — its goal, tasks, and preferred tone.
   An agent that monitors Jira sprints, identifies blockers, and posts a
   daily summary to the team Slack channel. Professional tone.
   
4. Click Generate. ELITEA drafts a Name, Description, Instructions, Welcome Message, and Conversation Starters for you.
5. Review and edit each field. Select any suggested toolkits or child agents to attach.
6. Click Create Agent to save.

---

​

How to select and configure Toolkits

Toolkits are integrations with external or ELITEA’s internal services that enhance your agent’s capabilities by allowing it to interact with various resources and perform specific tasks. In addition to toolkits, you can also add other Agents, MCPs (Model Context Protocol servers), and Pipelines to extend your agent’s functionality. You can add these resources to your agent during the creation process or edit them later. Adding Resources to Your Agent:

1. In the agent creation or editing interface, navigate to the TOOLS section.
2. Use the dedicated add buttons to attach resources. Each button opens its own searchable dropdown:
   • +Toolkit: Browse and select from available toolkits, or click “Create new” to create a new toolkit. See Toolkits.
   • +MCP: Browse and select Model Context Protocol servers, or click “Create new” to create a new MCP. See MCPs.
   • +Agent: Browse and select classic agents in the project to add as nested agents.
   • +Pipeline: Browse and select pipelines in the project. See Pipelines.
3. Each dropdown supports inline search and paginated scrolling to locate resources quickly.

​

SKILLS

Skills are reusable, Markdown-based instruction sets that you attach to an agent version to give it named, on-demand capabilities — without duplicating instructions across multiple agents. Attaching a Skill:

1. In the agent configuration form, scroll to the Skills accordion section.
2. Click the + Skill button to open the skill picker.
   • The button is disabled if the agent has not been saved yet (tooltip: “Save the agent first, then add skills”) or if the 5-Skill limit has been reached.
3. Search for the Skill by name, then click it to attach. It attaches using the Skill’s base version by default.

Changing the Attached Version: Each attached Skill card shows the active version name. Click the version name to open a Versions dropdown and select a different version. The attachment updates automatically. Removing a Skill: Hover over the Skill card and click the Remove skill (trash) icon. Confirm in the dialog.

---

​

INTERNAL TOOLS

ELITEA provides built-in internal tools that extend your agent’s capabilities without requiring external integrations. These tools enable your agents to perform specialized tasks directly within the platform. Available Internal Tools: How to Enable Internal Tools:

1. In the TOOLS section, scroll to the bottom to find the INTERNAL TOOLS subsection.
2. Each tool has its own toggle switch. If not all tools are visible, click Show all to expand the full list.
3. Enable the tools you need:
   • Attachments toggle
   • Image creation toggle
   • Data Analysis toggle
   • Planner toggle
   • Python sandbox toggle
   • Swarm Mode toggle
   • Smart Tools Selection toggle
   • ELITEA MCP Tools toggle

​

How to Create Instructions

The Instructions section is the core component where you define how your agent should behave and what tasks it should perform. This field serves as the foundational knowledge base that guides the AI model in understanding and processing your specific requests. How to Input Instructions

• Identify Key Information: Before entering data into the Instructions field, identify the essential details or instructions that the model needs to know to fulfill your request effectively. This could include the topic, specific terms, relevant background information, or the scope of the task.
• Enter the Details: In the Instructions field, clearly and concisely input the identified information. Ensure that the information is directly relevant to the task to maintain the agent’s focus and efficiency.

​

Mentions in Instructions

The Instructions field supports / mentions to reference tools and ~ mentions to reference Skills added to the agent directly in the instruction text. Inserted mentions are highlighted in the field so you can see at a glance which resources are referenced. How to use mentions:

1. Type / anywhere in the Instructions field to open the suggestion dropdown for toolkits, MCPs, agents, and pipelines.
2. Type ~ to open the suggestion dropdown for Skills attached to the agent.
3. Start typing after / or ~ to filter the list by name.
4. Use the arrow keys to navigate the list and Enter to select, or click an item directly.
5. For toolkits and MCPs, selecting an item opens a second phase to optionally pick a specific tool within that toolkit. Type to filter tools and select one, or press Enter or click the toolkit name without a tool to mention the whole toolkit.
6. Press Escape at any point to dismiss the dropdown without inserting a mention.

Mention format inserted into the instructions text:

​

VARIABLES

The Variables section appears automatically in the left configuration panel when your agent’s instructions contain one or more template placeholders in the {{variable_name}} syntax. Each placeholder is extracted and displayed as a separate input field, allowing you to define reusable values without hardcoding them directly in instructions. How Variables Work:

1. In the Instructions field, include placeholders using double curly braces: e.g., {{project_name}}, {{jira_ticket_id}}.
2. After saving, the Variables accordion appears below Instructions with an input field for each detected placeholder.
3. Enter the default or current value for each variable.
4. During execution, the placeholders are substituted with the provided values before the instructions are sent to the model.

​

EDITOR NOTES

The EDITOR NOTES section is a separate accordion in the left configuration panel.It is intended for internal documentation about the current agent version. Use this field to record implementation notes, review comments, rollout reminders, or other editor-facing context that should stay with the agent configuration.

• The content is not sent to the LLM, not included in chat or execution, and not used in monitoring

Example uses:

• Use this version for staging validation only.
• Pending toolkit credential rotation before production rollout.
• Draft variant created for testing alternative instructions.

​

WELCOME MESSAGE

The Welcome Message feature allows you to provide additional context and instructions that appear when users interact with your agent in the chat interface. This message helps set expectations and guide users on how to best utilize the agent. How to Add the Welcome Message:

1. Access the Welcome Message Section: In the agent creation or Configuration interface, navigate to the Welcome Message section.
2. Add the Welcome Message: Type the welcome message text in the input field.
3. Save the Configuration: After entering the desired text, ensure to save the changes to the agent. This action makes the configured welcome message available to users in the Chat section.

Using the Welcome Message: When users go to the Chat section of the agent, they will see the configured Welcome Message. It provides helpful context and instructions to guide their interaction with the agent.

​

CONVERSATION STARTERS

The Conversation Starter feature enables you to configure predefined prompts that help users quickly initiate specific types of interactions with your agent. These starters appear as clickable buttons in the chat interface, making it easy for users to get started. How to Add a Conversation Starter:

1. Access the Configuration Panel: Navigate to the Conversation Starter section in the agent creation or editing interface.
2. Add a Conversation Starter: Click the + icon to open the text input field where you can type the text you wish to use as a conversation starter.
3. Save the Configuration: After entering the desired text, ensure to save the changes to the agent. This action makes the configured conversation starter available for use.

Using a Conversation Starter: Initiate a Conversation: Go to the Chat section of the agent. Here, you will find the saved conversation starters displayed as clickable options. Click on the desired starter to automatically populate the chat input and execute the agent. By setting up conversation starters, you streamline the process of initiating specific tasks or queries, making your interactions with the agent more efficient and standardized.

​

INFORMATION

The Information section is an accordion at the bottom of the left configuration panel. It provides read-only metadata about the current agent and version.

​

Chat Panel

The right panel of the agent page contains a live chat panel that lets you test the agent without navigating away. Controls in the top bar of the chat panel: Context Budget widget (if context management is enabled): Appears above the chat panel after the first message. See Managing Context Budget for details.

​

How to Execute Agent

To execute the agent and get the output:

1. Configure the Agent: Ensure your agent is properly configured with clear instructions and any necessary toolkits.
2. Navigate to Chat: Access the agent’s chat interface by clicking on the agent from your agents list.

​

Selecting and Configuring the AI Model

Before sending a message, you can choose which AI model processes the agent and optionally fine-tune its parameters.

1. Click the model chip in the chat input bar to open the model selector dropdown.
2. Choose the desired model from the list (e.g., gpt-4o, gpt-5.1).
3. Click the Model Settings icon (⚙️) next to the model chip to adjust generation parameters.

Max Completion Tokens (All Models)

​

Managing Context Budget

When the context_manager secret is enabled at the project level, the Context Budget widget appears above the chat interface to help you manage conversation token usage effectively. What is Context Management? Context Management provides intelligent control over conversation token usage through automated message pruning and summarization. It helps maintain conversation continuity while staying within model token limits by automatically managing message history, generating summaries of older conversations, and preserving important messages. Accessing the Context Budget Widget

1. Ensure the context_manager secret is set to true in your project settings (Settings → Secrets)
2. Send the first message to initiate the agent conversation
3. The Context Budget widget appears above the chat panel after the first message
4. Monitor real-time token usage and management status
5. Click on the widget to expand and view detailed metrics

Widget Views

• Collapsed View: Shows essential token usage status with a color-coded indicator (Green: normal, Orange: high usage)
• Compact View: Displays pruning strategy, message count, and summaries count
• Expanded View: Provides comprehensive configuration options including:
  • Context Strategy & Token Management (pruning strategy, max tokens, preserve recent messages)
  • Summarization settings (enable/disable, summary parameters)
  • System Messages management

Key Features

• Real-time tracking: Monitor token consumption as your agent processes requests
• Automatic pruning: System removes older messages when context limit is reached
• Summarization: Generate summaries of conversation history to preserve context
• Message preservation: Configure how many recent messages to always keep

​

Starting and Interacting with Your Agent

Start Interaction: Begin your conversation by either:

• Clicking on a Conversation Starter (if configured) to use a predefined prompt
• Typing your question or command directly into the chat input field
• Using simple commands like “Go”, “Start Generating”, “Execute”, or “Run it” followed by clicking the Send button

Additional Interaction Features:

• Full Screen Mode: Increase the size of the output window for better visibility and focus. This mode can be activated to expand the output interface to the full screen.
• Continue the Dialogue: To keep the conversation going, simply type your next question or command in the chat box and click the Send icon.
• Copy the Output: Click the Copy to Clipboard icon to copy the generated text for use elsewhere.
• Regenerate Response: If the output isn’t satisfactory, click the Regenerate icon to prompt the Gen AI to produce a new response.
• Delete Output: To remove the current output from the chat, click the Delete icon.
• clear the Chat: For a fresh start or to clear sensitive data, click the Clean icon to erase the chat history.
• Like or Dislike the Output:
  • Click the Like icon if the output meets your expectations.
  • Click the Dislike icon if the output is unsatisfactory. Upon disliking, you will have the option to leave a comment explaining why the output did not meet your expectations. This feedback helps improve the system’s performance and relevance.

​

Voice Capabilities

The agent chat interface supports voice input and voice output, letting you dictate prompts and have agent responses read back to you.

---

​

Voice Input

Voice Input lets you dictate prompts directly into the agent chat input field. The transcribed text is inserted at the cursor position, so you can combine typed and spoken content in the same message. How to use Voice Input:

1. Click the microphone icon in the message input toolbar to start recording.
2. Speak your prompt. A live transcript appears in the input field as you talk — interim results update in real time.
3. Click the Stop (■) button to finish recording. The final transcript is committed and focus returns to the input field.

---

​

Text-to-Speech

Text-to-Speech reads agent responses aloud. A Read out button (megaphone icon) appears in the action bar below each agent response when the message contains speakable text. How to use Text-to-Speech:

1. Click the Read out (megaphone) button below an agent response to start playback.
2. The text is highlighted as it is read aloud.
3. A playback pill appears in the input area with Pause and Resume controls.
4. Click Pause to stop playback mid-sentence; click Resume to continue from where it left off.

---

​

Speaking Mode

Speaking Mode is a continuous, hands-free voice loop. Once activated, ELITEA automatically records your speech, sends your prompt after a pause, plays back the agent’s response, and then starts listening again — without any manual interaction between turns. How to activate Speaking Mode:

1. Leave the message input field empty.
2. Click the voicewave icon (shown in place of the Send button when the input is empty) to enter Speaking Mode.
3. A voicewave pill with an Exit (✕) button replaces the send button, indicating Speaking Mode is active.
4. Speak your prompt. After 3 seconds of silence, your message is automatically sent.
5. The agent response streams in and is read aloud via Text-to-Speech.
6. Once the response finishes playing, recording starts again for the next turn.
7. Click the ✕ button in the voicewave pill to exit Speaking Mode at any time.

---

​

Managing Agent Versions: Save, Create Versions, Publish and Manage

To optimally manage your agent, understanding how to save and create versions is crucial. Follow these guidelines to efficiently save your agent, create versions, and manage them. How to Save an Agent:

• To save your work on an agent for the first time, simply click the Save button. This action creates what’s known as the “latest” version of your agent.
• You can continue to modify your agent and save the changes to the “latest” version at any time by clicking the Save button again. If you wish to discard any changes made, you have the option to click the Discard button before saving.

Remember: The “latest” version represents the initial version you create. You can keep updating this version with your changes by saving them, without the need to create additional versions for your agent.

​

How to Create New Versions:

For instances where you need to create and manage different iterations of your agent:

1. Initiate a New Version: Start by clicking the Save As Version button.
2. Name Your Version: When saving your work, provide a version name that clearly identifies the iteration or changes made. Click Save to confirm your entry.

Best Practices for Version Naming:

• Length: Keep the version name concise, not exceeding 48 characters. This ensures readability and compatibility across various systems.
• Characters: Avoid using special characters such as spaces (” ”), underscores (”_”), and others that might cause parsing or recognition issues in certain environments.
• Clarity: Choose names that clearly and succinctly describe the version’s purpose or the changes it introduces, facilitating easier tracking and management of different versions.

Upon creating a new version of the agent, several options become available to you:

• Delete: Remove this version of the agent if it’s no longer needed.
• Execute: Run this specific version of the agent to see how it performs.
• Navigate Versions: Use the Version dropdown list to switch between and select different versions of the agent. This allows for easy comparison and management of various iterations.

​

Publishing an Agent Version

The Publish functionality allows you to make a specific Draft version of your agent available in Agents HUB — a shared library accessible to all users in your ELITEA installation. Publishing uses a guided three-step wizard with AI-powered automated validation that checks your agent before it can be submitted. How to Publish an Agent Version:

1. Open the agent and select the Draft version you want to publish using the version selector.
2. Click the three-dot menu (⋮) in the agent tab bar and select Publish.
3. In the Preparation step, enter a version name and accept the Publishing Terms, then click Continue.
4. The AI Validation step runs automatically — review the result (PASS, WARN, or FAIL). Address any critical issues if needed.
5. Click Publish to submit. On success, the version is immediately live in Agents HUB.

What Happens After Publishing:

• The version status updates to Published and appears in your Published tab.
• Community users can find and use the agent in Agents HUB without needing project access.
• You can unpublish at any time using the Unpublish option in the three-dot menu.

​

Exporting and Importing Agents

ELITEA allows you to export the currently selected version of an agent as a file and import it into another project, making it easy to back up, share, and migrate your work. Export and import are always version-specific — there is no option to export the entire agent with all its versions at once. Exporting an Agent Version:

1. Open the agent and select the version you want to export using the version selector.
2. Click the three-dot menu (⋮) in the toolbar.
3. Select Export. The file downloads automatically to your device.
   • If the selected version has no nested dependencies, a single .md file is downloaded.
   • If the selected version references other agents as toolkits, the export is a .zip file containing all dependencies.

Importing an Agent:

1. In the Agents dashboard, click the Import button in the toolbar.
2. Upload the .md or .zip file previously exported from ELITEA.
3. The Import Wizard displays entity cards for the main agent and any nested dependencies. Review configuration before confirming.
4. Click Import to complete. The agent will appear in the current project.

---

​

Forking Agents

Forking copies the currently selected version of an agent from one project to another within the same ELITEA environment — no file download required. Only one version is forked at a time; there is no option to fork the entire agent with all its versions. How to Fork an Agent Version:

1. Open the agent and select the version you want to fork using the version selector.
2. Click the three-dot menu (⋮) in the toolbar and select Fork.
3. In the Fork parameters wizard, choose a target project from the dropdown. No project is pre-selected; you must pick one manually.
4. Review the entity cards — the main agent card and any nested agent dependencies that will be forked automatically.
5. Click Fork. When complete, click Got it to navigate directly to the forked agent in the target project.

---

​

Viewing Agent History

The Run History panel provides a complete audit trail of all past executions of your agent. How to access Run History:

1. Open any saved agent.
2. In the right panel (embedded chat area), click the clock icon (🕐) in the top bar. The tooltip reads “View run history”.
3. The agent page is replaced by the two-panel Run History display. Click the ✕ (Close) button in the top-left to return.

What Run History shows:

• Review past conversations: View complete chat histories from previous agent runs
• Track performance: Monitor execution duration across different versions
• Debug issues: Replay conversations to identify where problems occurred
• Compare versions: See how different agent versions performed with the same inputs
• Audit trail: Maintain records of all agent interactions for compliance purposes

Run History layout: The Run History panel uses a two-panel layout:

• Left panel — sortable list of all past runs. Click any column header to sort by that column:

  Column	Description
  Date	Timestamp of the run in dd-MM-yyyy, hh:mm a format
  Version	Agent version used for the run
  Duration	Total execution time of the run
  Click a row to load the conversation replay in the right panel.
  Each row also has a three-dot menu (⋮) with the following actions:
  Action	Description
  ---	---
  Share link	Copies a direct URL to this run to the clipboard
  Delete	Permanently removes this run record (with confirmation prompt)
  Restore conversation	Pre-populates the embedded test chat with the selected run’s message history. Note: restores chat history only — agent configuration, instructions, and tool settings are not reverted.
  The list supports infinite scroll — additional runs load automatically as you scroll down.

• Right panel — full conversation replay of the selected run.

​

Best Practices

Write Clear Instructions

Provide specific, actionable instructions that clearly define your agent’s behavior and goals. Use examples and step-by-step guidance.

Test Your Agent Thoroughly

Test your agent with various inputs before deploying it. Use conversation starters to verify expected behavior.

Use Version Management

Create named versions for significant changes. This allows you to roll back if needed and compare different iterations.

Monitor Agent Performance

Regularly review agent history to identify issues, track performance, and understand how users interact with your agent.

Keep Toolkits Updated

Ensure connected toolkits have valid credentials and up-to-date configurations. Test toolkit integrations regularly.

Enable Context Management

For long conversations, enable context management to optimize token usage and maintain conversation quality.

---

​

Troubleshooting

Agent Not Responding or Slow

Check your AI model selection and settings. High reasoning levels or large token limits may slow responses. Verify toolkit connections are working.

Unexpected or Incorrect Outputs

Review agent instructions for clarity. Ensure toolkits are properly configured. Check conversation history to understand context issues.

Toolkit Integration Failures

Verify credentials are valid and not expired. Test the toolkit independently using the Test Settings panel. Check network connectivity and permissions.

Cannot Publish Agent

Ensure you have publishing permissions. Check that all required fields are completed. Review moderation guidelines if submission was rejected.

Version Not Available

Verify the version was saved successfully. Check permissions if trying to access another user’s version. Refresh the page and try again.

Agent Not Appearing in List

Clear search filters and check status filters. Verify you’re in the correct project. Refresh the agents dashboard.

For further assistance, contact your platform administrator.

​

Support Contact

If you encounter issues not covered in this guide or need additional assistance with Agent management, please refer to Contact Support for detailed information on how to reach the ELITEA Support Team.

---