Agents

Agents are AI assistants configured to perform specific tasks. In AI Cockpit Lens, each agent acts as a processing unit with its own system prompt, AI provider and model, user message template, context variables, tools, and linked knowledge bases. Agents can support reasoning, code, chat, documentation, testing, requirements analysis, architecture review, and custom workflows.

When to use

• Use agents to create assistants with a clear objective.
• Use the agents list to review existing assistants and open their details.
• Use the playground to test an agent before adding it to important workflows.
• Use agents as the reusable building blocks that compose workflow steps.

Before you start

• Confirm the active organization.
• Define the purpose of the agent before creating it.
• Prepare the instructions that should guide the agent's behavior.
• Confirm which AI model options are available for the organization.
• Prepare any context variables, tools, or knowledge bases the agent should use.

Step by step: browse agents

1. Click Agents in the main navigation.
2. Use search to filter by name or description.
3. Sort by name or date when needed.
4. Open an agent to review its details.

Step by step: create an agent

1. Click Create Agent.
2. Fill in the basic information: name, type, and description.
3. Choose the AI provider and model available for your organization.
4. Enter the system prompt that defines how the agent should behave.
5. Add the user message template when the agent needs a structured input format.
6. Configure context variables, tools, and linked knowledge bases when available.
7. Adjust settings such as timeout, retry attempts, and streaming when available.
8. Click Save Agent.

Understand the main fields

Field	What it means
Name	Display name shown in lists and details.
Type	Category that describes the agent's purpose.
Description	Short explanation of what the agent does.
Provider and model	AI service and model used by the agent.
System prompt	Main instruction that guides the agent's behavior.
User message template	Structure used to assemble the request sent to the agent.
Context variables	Dynamic values included when the agent receives a request.
Tools	External capabilities the agent can call when enabled.
Knowledge bases	RAG sources the agent can consult for organization or project context.
Timeout	Maximum expected response time.
Streaming	Allows the answer to appear progressively when supported.

Step by step: test in the playground

1. Open the details of an active agent.
2. Click Open Playground.
3. Type a message in the input field.
4. Click Execute Agent.
5. Wait for the response.
6. Review status, usage, cost, and duration when those details are shown.

Step by step: edit or delete an agent

1. Open the agent details or use the list action.
2. Click Edit to change the configuration.
3. To remove the agent, choose Delete and confirm the action.

caution

Deleting an agent can affect workflows that depend on it.

Expected result

You should have an agent that is named clearly, configured with the right instructions, and tested before being reused in workflows.

Tips and common issues

• If the agent output is not aligned with the task, review the system prompt and user message template.
• If execution fails, check whether the selected provider and model are available for the organization.
• If the agent is used by workflows, review those workflows before editing critical settings.