Skip to main content
Insight Flow splits content processing into multiple stages. Each stage can choose its own execution provider so you can balance cost, speed, and output quality.

​
Stage Model

  • filter: remove low-value or irrelevant items
  • keywords: extract keywords
  • global_summary: generate a summary across multiple items
  • report: generate the final report body

​
Available Providers

  • rule: rule-based execution
  • llm_openai: OpenAI-compatible model execution
  • llm_codex: Codex-compatible model execution
Notes:
  • global_summary and report only support LLM providers
  • Check defaults through GET /api/v1/monitors/ai-routing/defaults

​
How to Use It

  1. Configure and test providers on the Providers page
  2. Create or edit a monitor
  3. Choose a provider for each stage
  4. Run a small monitor first to compare latency and output quality

​
Common Patterns

  • Cost-first: filter=rule, keywords=rule, global_summary=llm_openai, report=llm_openai
  • Single-model: all LLM stages use llm_codex
  • Hybrid: rules for lightweight stages, LLMs for long-form output