Skip to main content

Agents

The Agent node is the intelligence engine of your workflow. It represents a Large Language Model (LLM) configured to perform specific reasoning tasks.

To configure an agent, click on any Agent Node in your canvas to open its settings panel.

info

The effectiveness of an agent depends almost entirely on its System Prompt—the instruction that defines its persona and goals.

Agent Configuration​

1. System Prompt​

The core instruction that tells the agent who it is and what it should do.

Description: Defines the behavior, boundaries, and tone of the agent. A precise prompt prevents hallucination and ensures accurate outputs.

Example:

"You are a Senior DevOps Engineer. You only write Python scripts that are production-ready, error-handled, and commented."

Best For: Specializing agents for distinct roles (e.g., "Copywriter", "Data Analyst", "Coder").

2. Model Selection​

Choose the specific AI model that powers this agent node.

Options:

  • High Reasoning (e.g., GPT-4o, Claude 3.5 Sonnet): Capable of complex planning, coding, and nuanced understanding.
  • Fast & Efficient (e.g., GPT-4o-mini): Optimized for speed and lower cost.

Best For: Balancing performance and budget. Use "High Reasoning" for the main brain and "Fast" for simple summarization tasks.

3. Temperature​

Controls the randomness and creativity of the model's output.

Configuration:

  • Low (0.0 - 0.3): Deterministic and precise. The model chooses the most probable answer every time.
  • High (0.7 - 1.0): Creative and diverse. The model explores novel and varied responses.

Best For:

  • Low Temp: Data extraction, coding, SQL generation.
  • High Temp: Brainstorming, creative writing, marketing copy.

4. Tool Augmentation​

Connects the agent to external services and APIs.

Description: Allows the LLM to execute actions in the real world instead of just generating text. The agent intelligently decides when to call a tool based on the user's request.

Popular Integrations:

  • Productivity: Gmail, Google Calendar, Google Sheets, Notion, Slack
  • Development: GitHub, Jira
  • Search & Data: Tavily, DuckDuckGo
  • Social: Twitter/X, LinkedIn

Best For: Agents that need to fetch real-time data or perform actions (e.g., "Check my calendar and schedule a meeting").

5. Knowledge Base - RAG​

Give your agent "long-term memory" by connecting it to your custom documents.

Description: instead of relying only on its training data, the agent can "read" your uploaded PDFs, CSVs, or text files to answer specific questions.

How it works: This uses Retrieval-Augmented Generation (RAG). The agent searches your documents for relevant info before generating an answer.

Learn More

For a step-by-step guide on uploading documents and managing your data, see the dedicated Knowledge Base guide.