Reasoning

AI-powered reasoning over live media streams using Plan + Evaluate architecture

Component Configuration

Option Description Required Environment override

id

A unique identifier for the component instance

displayName

The name shown in the Studio UI of this component

mode

Where the evaluation design comes from: the AI planner designs it from your query, you draw it as a graph, or you paste a previously copied spec

analysisIntervalSecs

Seconds between LLM prompts (Generative: analysis interval, Live: periodic prompt interval) (default: 2)

contextDocuments

File paths to reference documents (relative to data directory, e.g. rules.md) (default: [])

notes

Optional notes about this component

Mode (AI planner)

Option Description Required

kind

"planning"

api

Generative analyses frames on an interval through Plan/Evaluate; Live streams continuously over the provider's realtime API (no local analysis) (default: "generative")

query

Natural-language query for the planning LLM

autoAccept

Auto-accept first proposed spec (default: true)

planningProvider

LLM provider for planning (spec generation)

evaluationProvider

LLM provider the accepted spec is evaluated with

designFlow

Let the planner design the local-analysis flow itself — tools, gates, cadence — from the palette below (on by default); artifacts it cannot supply surface as requirements on the spec. Off = a fixed default tool trio (audio energy, VAD, scene change) the planner may gate on. (default: true)

toolPalette

Tools the planner may design with (empty = all) (default: [])

events

Events the LLM can raise (default: [])

Mode (Manual graph)

Option Description Required

kind

"graph"

graph

The local-analysis flow: tool nodes whose gate references form the edges, plus an optional LLM terminal step (default: {"tools":[]})

Mode (Baked spec)

Option Description Required

kind

"baked"

specJson

Paste a spec here (JSON or YAML from 'Copy Spec' button); whether it targets the Generative or Live API is detected from its shape

evaluationProvider

Only used when the pasted spec includes an LLM stage — providers and their keys never travel inside a spec. A tools-only spec needs no provider.

restorePlanning

If the pasted spec was produced by 'Copy Spec', restores the original planning configuration (api, query, provider, events) into the form (default: null)

Tags: ai reasoning llm plan evaluate gemini openai claude detection analysis