57.1 Architecture Generation Router Service

Documentation for Sessions

57.1 Architecture: GenerationRouterService for Unified AI Generation

Date: January 20, 2026 Type: Architecture Decision Confidence: 95% (Certain)

Context

The "Generate Widget" button in the AI Panel's widget tab was not working - no logs appeared in backend-api or backend-ai-api. Investigation revealed that IntelligenceCoordinatorService.processCompletion() was returning placeholder code instead of parsing actual AI responses.

Decision

Create GenerationRouterService as a unified entry point for all AI generation requests that routes directly to the correct backend-api endpoints.

Options Considered

Fix only IntelligenceCoordinatorService.processCompletion() - Quick fix but doesn't address fragmented generation paths
Create GenerationRouterService + fix processCompletion() ✅ CHOSEN - Clean abstraction + backward compatibility
Bypass all frontend services and call backend directly from components - Too scattered, no reusability

Implementation

GenerationRouterService Routes

Generation Type	Backend Endpoint
widget	POST /api/ai/generate/widget
module	POST /api/ai/generate/module
image	POST /api/ai/images/generate
video	POST /api/ai/videos/generate
code	POST /api/ai/chat/generate-code
chat	POST /api/ai/chat

Key Features

Unified UnifiedGenerationRequest interface for all types
Normalized GenerationResult response format
History tracking with generationHistory signal
Loading state management
Error handling with proper logging

processCompletion() Fix

The fix includes:

Regex patterns to extract TypeScript, HTML, SCSS, and spec code blocks
Confidence scoring based on code completeness
Fallback generation for when parsing fails
Extraction of imports and dependencies from generated code

Alignment with Session 56 Decisions

Session 56 decided to integrate gograb-modules' PhaseExecutor into backend-ai-api. This aligns because:

GenerationRouterService is an abstraction layer - When PhaseExecutor is integrated, we only need to change the routing targets
We're not creating a new separate service - Following the "integrate into existing" principle
Backend endpoints are already production-ready - Leveraging existing infrastructure

Files

libs/libs-ai/gograb-ai/src/lib/services/generation-router.service.ts (NEW)
libs/libs-ai/gograb-ai/src/lib/services/intelligence-coordinator.service.ts (MODIFIED)

Consequences

Positive:

Widget generation now routes to backend correctly
Single entry point for all generation types
Easy to add PhaseExecutor integration later

Neutral:

Two ways to generate (old IntelligenceCoordinator, new GenerationRouter)
This is intentional for backward compatibility

Negative:

None identified