OpenQCore · Multimodal Stack

One Stack.
Four Modalities. Production Control.

Multimodal Stack unifies text, voice, image, and video in one execution architecture with routing intelligence, provider fallback chains, and stable response contracts.

Text + Voice + Image + VideoDecision RouterTimeout MatrixFallback ChainsStreaming / Non-StreamingUnified Output Contract

Decision Matrix

Intent-first routing across modalities.

PulseDecisionRouter classifies user intent and dispatches execution to the right lane: chat, image generation/edit/analyze, video generation, or tool call.

Modality

Text

Live Route Preview

Trigger pattern

default conversational intent

Route

chat

Engine / Provider

Pulse + Chat Runtime

Provider Timeouts

Latency budgets by workload type

Image

120s

Fast generation budget

Video

300s

Long-running render window

OCR

45s

Document extraction budget

Identity

90s

KYC flow budget

Fraud

45s

Risk analysis budget

Fallback Chains

Provider reliability routing

Image: openai

openai
replicate

Image: replicate

replicate
openai

Video

ltx_video

OCR

google_vision
PrimaryFallback HopExecution Target

Execution Paths

Two delivery modes, one orchestration core.

Streaming path

  • Engine returns async generator for token/event flow
  • StreamProcessor wraps SSE delivery
  • Timeout and lock controls remain active
  • Final response closes with deterministic cleanup

Non-streaming path

  • Single execution with timeout + circuit breaker
  • Structured payload assembly (content/images/files/videos)
  • Usage accounting and duration metrics
  • ResponseBuilder returns canonical JSON response

Unified Output Contract

Single response envelope for all modalities.

The runtime returns a consistent contract that keeps frontend rendering predictable across mixed text, file, image, and video outputs.

images

Generated or analyzed image artifacts mapped in a consistent array.

files

Attachments, OCR files, and generated documents unified for client rendering.

videos

Video generation outputs exposed through the same response payload model.

Ship multimodal products with runtime-level confidence.

Build once across text, voice, image, and video with production controls for routing, fallback, observability, and stable output contracts.