Skip to main content

​
Overview

Standard Mode is the simplest execution mode for the AI Agent Node. The node assembles a prompt from the system message, conversation history, and user input, makes a single LLM call, and returns the response. There are no intermediate reasoning steps, no tool calls, and no self-critique loops. This is the default mode and should be your starting point for any new workflow. Upgrade to a more complex mode only when Standard does not meet your quality requirements.

​
How It Works

1

Assemble Prompt

The node combines the system prompt, conversation history (if memory is enabled), and the current user message into a single prompt.
2

Call LLM

The assembled prompt is sent to the configured model. The response streams token-by-token via SSE llm_token events.
3

Return Response

The complete response is written to the workflow context and delivered to the user.

​
Configuration

{
  "type": "ai-agent-node",
  "config": {
    "agent_mode": "standard",
    "model": "gpt-4o",
    "system_prompt": "You are a helpful customer support assistant for Nadoo AI. Answer questions clearly and concisely.",
    "temperature": 0.7,
    "max_tokens": 4096
  }
}
ParameterTypeDefaultDescription
agent_modestring—Must be "standard"
modelstring—Model identifier (e.g., "gpt-4o", "claude-sonnet-4-20250514", "ollama/llama3")
system_promptstring""Instructions that define the agent’s behavior and persona
temperaturefloat0.7Controls randomness (0 = deterministic, 2 = highly random)
max_tokensint4096Maximum tokens in the response

​
Additional Model Settings

{
  "agent_mode": "standard",
  "model": "gpt-4o",
  "system_prompt": "You are a translator.",
  "temperature": 0.3,
  "max_tokens": 2048,
  "top_p": 0.9,
  "frequency_penalty": 0.2,
  "presence_penalty": 0.1,
  "stop_sequences": ["---"]
}
ParameterTypeDefaultDescription
top_pfloat1.0Nucleus sampling threshold
frequency_penaltyfloat0.0Penalize frequently appearing tokens (-2.0 to 2.0)
presence_penaltyfloat0.0Penalize tokens that have appeared at all (-2.0 to 2.0)
stop_sequencesstring[][]Sequences that cause the model to stop generating

​
SSE Events

Standard mode emits the following events during execution:
EventWhenPayload
node_startedNode begins execution{ node_id, node_type }
llm_tokenEach token is generated{ token, node_id }
llm_finishedLLM generation completes{ node_id, total_tokens }
node_finishedNode completes{ node_id, status }
Standard mode does not emit llm_thinking, llm_tool_call, or agent_reflection events.

​
Use Cases

Simple Q&A

Answer user questions based on the system prompt and conversation history. No external data or tools needed.

Content Generation

Generate blog posts, emails, marketing copy, or other text content from a prompt.

Translation

Translate text between languages. Set a low temperature (0.2-0.3) for consistent, accurate translations.

Summarization

Summarize documents, articles, or conversation threads. Works well when the full text fits within the context window.

Classification

Classify text into categories (sentiment, topic, language). Pair with a Condition Node downstream to branch on the result.

Formatting & Extraction

Reformat data, extract structured fields from unstructured text, or convert between formats (JSON, CSV, Markdown).

​
Example: RAG with Standard Mode

Standard mode works well for RAG workflows where the knowledge retrieval is handled by upstream nodes: The Search Knowledge Node retrieves relevant context, and the AI Agent Node in Standard mode uses that context to generate a response. No complex reasoning or tool use is needed because the retrieval is already done.

​
System Prompt for RAG

You are a knowledge base assistant. Answer the user's question based on the following retrieved context. If the context does not contain relevant information, say "I don't have information about that."

Context:
{{search_results}}

​
Performance Characteristics

MetricStandard Mode
LLM calls per execution1
LatencyLowest (single round-trip)
Token usageLowest (no overhead from reasoning, reflection, or tool calls)
Quality ceilingModerate (limited by single-pass generation)
Standard mode is typically 2-5x faster and 2-5x cheaper than multi-step modes like ReAct or Tree of Thoughts. Always benchmark Standard mode first before moving to a more complex strategy.

​
When to Upgrade

Consider switching to a more advanced mode when:
  • Accuracy on reasoning tasks is poor — Try Chain of Thought for step-by-step reasoning.
  • The agent needs to call tools — Use ReAct or Function Calling mode.
  • Output quality needs iterative improvement — Use Reflection for self-critique.
  • Multiple approaches should be explored — Use Tree of Thoughts for parallel reasoning.

​
Best Practices

In Standard mode, the system prompt is your primary lever for controlling output quality. Be specific about the desired format, tone, length, and any constraints.
Set temperature to 0.0-0.3 for translation, extraction, and factual Q&A. Use 0.7-1.0 for creative tasks where variety is desirable.
Use Search Knowledge, Database, or Variable Nodes upstream to provide the AI Agent with all the context it needs. Standard mode excels when the input is well-prepared.
Avoid using very large max_tokens values for tasks that need short answers. This wastes budget on unused capacity and can lead to unnecessarily verbose responses.

​
Next Steps

Chain of Thought

Add step-by-step reasoning for complex problems

ReAct Mode

Enable tool use with iterative reasoning

AI Agent Node

Full configuration reference for all modes

Strategies Overview

Compare all 6 execution modes