Skip to content

Examples

Node IDs in these diagrams are workflow node names. The template syntax in labels ({{ agent_executor.response }}) is how downstream FlowDrop nodes reference this node's outputs.

Run an agent and show the answer

flowchart TD
    input["task_input (text input)"] -->|text to task_description| agent["agent_executor<br/>agent_id: my_agent<br/>create_directly: true"]
    agent -->|response| output["output (text output)"]

Continue a conversation

Build the history upstream and connect it to chat_history. Set create_directly to false while the agent is still gathering requirements, so it proposes changes instead of making them.

flowchart TD
    history["build_history (data processor)"] -->|output to chat_history| agent
    message["chat_input"] -->|text to task_description| agent["agent_executor<br/>agent_id: conversational_agent<br/>create_directly: false"]
    agent -->|response| reply["chat_output"]

Branch on the result

flowchart TD
    agent["agent_executor<br/>agent_id: content_creator"] -->|success| check{"check_success<br/>(if condition)"}
    check -- true --> ok["success_handler<br/>{{ agent_executor.structured_output.created_contents }}"]
    check -- false --> refused["refused_handler<br/>{{ agent_executor.error_message }}"]
    agent -->|error port| failed["log_error (logger)"]

The false branch handles agents that refuse the task. The error port handles runtime failures. See Error handling.

Restrict a RAG agent to one search index

Use the RAG Search Config node from FlowDrop AI Search to build tool usage limits for a search index, and connect its output to tool_usage_limits. The agent must be config-based.

flowchart TD
    rag["rag_config (RAG Search Config)<br/>index: content"] -->|tool_usage_limits| agent["agent_executor<br/>agent_id: rag_agent"]
    question["chat_input"] -->|text to task_description| agent
    agent -->|response| out["chat_output"]

Pick the model per workflow

Connect a string such as anthropic__claude-sonnet-4-5 to llm_model from an upstream node to run the same agent on a different provider or model. Leave it unconnected to use the agent's own settings.