CascadeFlow Agent icon

CascadeFlow Agent

CascadeFlow AI Agent with drafter/verifier orchestration, tool routing, domain routing, and trace metadata.

CascadeFlow Agent

AI-generated

Summary

Send a user input text message to the CascadeFlow Agent and receive an AI-generated response with configurable quality validation, complexity handling, domain-specific cascading, and operational constraints.

Inputs

  • System Message — Optional system prompt setting the overall behavior and context of the agent.
  • Text (required) — The user message to send to the agent, defaulting to the 'chatInput' field from a prior Chat Trigger node.
  • Quality Threshold — Minimum overall quality confidence score (0 to 1) that the agent's drafter response must meet to be accepted.
  • Use Complexity Thresholds — When enabled, applies different confidence thresholds based on detected query complexity levels from trivial to expert.
  • Trivial Threshold — Minimum confidence score required for trivial complexity queries.
  • Simple Threshold — Minimum confidence score required for simple complexity queries.
  • Moderate Threshold — Minimum confidence score required for moderate complexity queries.
  • Hard Threshold — Minimum confidence score required for hard complexity queries.
  • Expert Threshold — Minimum confidence score required for expert complexity queries.
  • Enable Alignment Scoring — Enables scoring of query-response alignment to improve validation accuracy.
  • Enable Complexity Routing — Enables routing of complex queries directly to the verifier model to optimize response quality.
  • Enable Tool Call Validation — Enables validation of drafter-generated tool calls (e.g., JSON syntax, schema, safety) before execution; invalid tool calls are regenerated by the verifier.
  • Max Tool Iterations — Maximum number of iterations for tool-call validation and regeneration.
  • Tool Routing Rules — Allows overriding of routing behavior per tool, specifying whether tool calls use the default cascade or go directly to the verifier.
  • Enable Domain Cascading — Enables domain-specific cascading logic for routing queries based on detected domains like code, legal, or medical.
  • Enable Domain Verifiers — Activates separate domain-specific verifier input ports to override the global verifier per domain.
  • Enable Code Domain — Add a domain input for programming and coding related queries.
  • Enable Data Analysis Domain — Add a domain input for data analysis and statistics queries.
  • Enable Structured Output Domain — Add a domain input for structured data extraction (e.g., JSON, XML).
  • Enable RAG (Retrieval) Domain — Add a domain input for retrieval-augmented generation and document Q&A.
  • Enable Conversation Domain — Add a domain input for chat and multi-turn conversations.
  • Enable Tool Calling Domain — Add a domain input for function calling, tool use, and API interactions.
  • Enable Creative Writing Domain — Add a domain input for creative writing tasks.
  • Enable Summarization Domain — Add a domain input for text summarization tasks.
  • Enable Translation Domain — Add a domain input for language translation tasks.
  • Enable Mathematics Domain — Add a domain input for mathematical reasoning and calculations.
  • Enable Science Domain — Add a domain input for scientific knowledge and research.
  • Enable Medical Domain — Add a domain input for healthcare and medical queries.
  • Enable Legal Domain — Add a domain input for legal documents and regulations.
  • Enable Financial Domain — Add a domain input for finance and accounting queries.
  • Enable Multimodal Domain — Add a domain input for image, audio, and video understanding.
  • Enable General Domain — Add a general-purpose domain input as a fallback.
  • Domain-Specific Settings — Configure per-domain quality thresholds and temperature settings to override global defaults.
  • Harness Mode — Select the operational harness mode: off (disabled), observe (telemetry only), or enforce (stop agent when limits exceeded).
  • Budget (USD) — Maximum budget in USD for agent operation; zero means unlimited.
  • Max Tool Calls — Maximum allowed tool call count; zero means unlimited.
  • Max Latency (Ms) — Maximum cumulative latency in milliseconds; zero means unlimited.
  • Max Energy — Maximum energy proxy units for operation cost; zero means unlimited.
  • Compliance — Compliance policy to enforce model allowlists (e.g., GDPR, HIPAA, PCI).
  • KPI Weights — Weights to optimize agent scoring dimensions (quality, cost, latency, energy), normalized automatically.

Output shape

A single JSON item containing the AI-generated response text and metadata including generation details and cascadeflow diagnostics.

The output includes one generation with the text result and associated message metadata. The operation streams partial responses when used in streaming mode. The cascade handles complexity-based routing, quality validation, tool-call validation, and fallback to verifier model if needed.

Examples

Example 1: Chatbot response with default settings

User message is provided; system message is optional; default quality threshold and complexity handling enabled; domain routing disabled; no operational harness limits; outputs the generated text response.

Links

Discussion