Agents

The Agents feature enables users to create and manage custom AI agents tailored to specific use cases, without needing to write code.

Key Capabilities

Configuration Panel On the left side there is the configuration panel of the agent, where the user can do the following:

Name - Set/edit the agent's display name.
Agent Type - Simple: Fast, single-step responses; Dee: Multi-step reasoning with planning.
Description - Optional, describe what the agent does.
Model - Select the underlying LLM that powers the agent, with additional model settings accessible via the adjacent settings icon.
System Prompt - Define the agent's behavior/persona in a free-text field (up to 8000 characters, with live character count).
Tools - Add tools/integrations the agent can use to perform actions or fetch external data.
Knowledge - Attach a knowledge base as a source of contextual/reference data for the agent.
Skills - Configure additional skills/capabilities for the agent.

Playground On the right side there is the playground: a live chat interface to test the agent's responses in real time before publishing, using the configured model, prompt, tools, and knowledge base.

Chat window with message history area (shows agent name and prompt "Send a message to test your agent" when empty)
Text input field ("Message the agent...") to send test messages, supporting Enter to send / Shift+Enter for new line
Option to add/attach additional context to the conversation (via the "+" icon)

Agent Lifecycle

Agents are saved automatically (with timestamp, e.g. "Saved Today 11:45"). Agents exist in Draft state until explicitly published via the Publish button.

Per-Agent Insights (tabs)

Playground - build/test the agent
Traces - view execution logs/traces of agent runs
Metrics - performance/usage metrics
Evaluations - quality/accuracy evaluation results

Key Capabilities

Configuration Panel On the left side there is the configuration panel of the agent, where the user can do the following:

Name - Set/edit the agent's display name.

Agent Type - Simple: Fast, single-step responses; Dee: Multi-step reasoning with planning.

Description - Optional, describe what the agent does.

Model - Select the underlying LLM that powers the agent, with additional model settings accessible via the adjacent settings icon.

System Prompt - Define the agent's behavior/persona in a free-text field (up to 8000 characters, with live character count).

Tools - Add tools/integrations the agent can use to perform actions or fetch external data.

Knowledge - Attach a knowledge base as a source of contextual/reference data for the agent.

Skills - Configure additional skills/capabilities for the agent.

Chat window with message history area (shows agent name and prompt "Send a message to test your agent" when empty)

Text input field ("Message the agent...") to send test messages, supporting Enter to send / Shift+Enter for new line

Option to add/attach additional context to the conversation (via the "+" icon)