Skip to content

codebase-context.md

Latest commit

 

History

History
54 lines (44 loc) · 2.52 KB

codebase-context.md

File metadata and controls

54 lines (44 loc) · 2.52 KB
Raw

LLM Adapter Kit - Codebase Context

Project Overview

Universal TypeScript library for interacting with multiple LLM providers through a unified interface. Supports 600+ models across 8 major providers with built-in cost calculation, streaming, and advanced features.

Current Task: Google Gemini API Updates

Status: In Progress - Updating to latest @google/genai SDK patterns Priority: Focus on Imagen models for image generation, latest Gemini 2.0 models for text

Architecture

  • Base Structure: BaseAdapter pattern with provider-specific implementations
  • Testing: Real API call validation with comprehensive test suite
  • Type Safety: Full TypeScript with proper type definitions

Google Adapter Structure

  • GoogleAdapter.ts - Text generation with Gemini models
  • GeminiImageAdapter.ts - Image generation with Imagen models
  • GoogleModels.ts - Model specifications and pricing

Key Dependencies

  • @google/genai: ^1.9.0 - Latest Google Gen AI SDK
  • TypeScript 5.4.5
  • Jest for testing

Recent Changes

  1. PREPARE Phase: Analyzed current implementation and latest API docs
  2. ARCHITECT Phase: Planned separation of text vs image generation
  3. CODE Phase: Currently implementing updates to use latest API patterns

Implementation Plan

  1. ✅ Update GoogleAdapter.ts - latest text generation API
  2. ✅ Update GeminiImageAdapter.ts - Imagen API pattern
  3. ✅ Update GoogleModels.ts - latest Imagen models and pricing
  4. ✅ Update tests - real API validation with Imagen

API Patterns Implemented

  • Text: Updated to use ai.models.generateContent() pattern
  • Images: Implemented ai.models.generateImages() with response.generatedImages[].image.imageBytes
  • Focus on Imagen models: imagen-4.0-generate-preview-06-06, imagen-4-ultra

Completed Updates

  • ✅ GeminiImageAdapter: Updated to use exact user-specified API pattern
  • ✅ GoogleAdapter: Updated to latest SDK patterns for text generation
  • ✅ GoogleModels: Kept existing text models, updated image model specs
  • ✅ Tests: Updated integration tests for Imagen 4 and 4 Ultra with realistic scenarios

Key Features Implemented

  • ✅ Imagen 4 support with multiple image generation (1-4 images)
  • ✅ Imagen 4 Ultra support (single image, higher quality)
  • ✅ Proper aspect ratio control (square, portrait, landscape, widescreen, fullscreen)
  • ✅ Person generation control (allow/block)
  • ✅ Cost calculation ($0.04 per image for Imagen 4, $0.06 for Ultra)
  • ✅ Validation for model-specific constraints
  • ✅ Base64 image response handling