Skip to content

CHANGELOG.md

Latest commit

 

History

History
221 lines (190 loc) · 10.3 KB

CHANGELOG.md

File metadata and controls

221 lines (190 loc) · 10.3 KB
Raw

Changelog

All notable changes to AI Studio are documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

1.1.0 - 2026-02-05

Added

  • API v1 Platform — REST API with OpenAPI spec for external integrations. Endpoints for assistants, context graph, conversations, and jobs. Authenticated via API key, OAuth JWT, or session.
  • OAuth2/OIDC Provider — Authorization Code Flow with PKCE. Admin UI for client registration. KMS-signed JWTs in production. Access tokens (15min), refresh tokens (24hr), ID tokens.
  • MCP Server — Model Context Protocol server exposing 5 tools (search_decisions, capture_decision, list_assistants, execute_assistant, get_decision_graph) over streamable HTTP transport.
  • Decision Framework — Structured decision capture with context, alternatives, evidence, and outcomes. Graph-based relationships between decisions. Full-text and semantic search.
  • K-12 Bedrock Guardrails — Amazon Bedrock content filtering and PII protection. DynamoDB-backed configuration. SNS real-time alerts. Detect-only mode by default.
  • API Key Management — User-managed API keys with sk- prefix. Argon2id hashing. Per-key scopes, rate limiting (60 req/min), usage analytics. Max 10 keys per user.
  • User Settings Page — Consolidated settings UI with API keys tab, preferences, and profile management.
  • Assistant Architect Improvements — Conversation continuity across executions. Provider-based model filtering. Enhanced SSE event streaming.
  • Dashboard Redesign — Updated layout with usage statistics and quick actions.
  • Tutorials Library — In-app tutorial content for onboarding educators and students.
  • Chart Visualization — Recharts-based data visualization in conversations.
  • Local Docker Development — npm run db:up, npm run dev:local, npm run db:seed for fully local development without AWS dependencies.
  • Integration Guides — New docs for MCP, OAuth, and API v1 quickstart.

Changed

  • Database ORM — Complete migration from RDS Data API to Drizzle ORM with postgres.js driver. Type-safe queries via executeQuery/executeTransaction. Connection pooling (max 20 per container).
  • Next.js 16 — Upgraded from Next.js 15 to 16.1.6.
  • AI SDK v6 — Upgraded from Vercel AI SDK v5 to v6.0 with updated provider packages.
  • React 19 — Updated to React 19.2.4.
  • Lambda Power-Tuning — 66% memory reduction (3GB to 1GB default) via AWS Lambda Power Tuning.

Fixed

  • SSE connection error log noise (#747)
  • Model-not-found errors (#746)
  • Timestamp conversion issues (#745)
  • Presigned URL refresh for expired S3 URLs (#737)
  • Guardrails false positives (#727, #742) — Disabled PROMPT_ATTACK filter, set topics to detect-only
  • Message count reconciliation (#719)
  • React Compiler violations (#461)

Infrastructure

  • GuardrailsStack — New CDK stack: Bedrock Guardrails + DynamoDB config table + SNS alert topic
  • Aurora Auto-Pause — Dev environments scale to 0 ACU when idle (~$44/month savings)
  • ECS Spot Instances — 70% cost savings on non-critical workloads
  • VPC Consolidation — Reduced from 2 VPCs to 1 shared VPC across all stacks

Breaking Changes

  • documents.conversation_id column type changed to UUID
  • chatEnabled column removed from relevant tables
  • OAuth token response includes new fields (id_token, updated scope format)
  • API key format standardized to sk- prefix

Dependencies

Package Version
next 16.1.6
ai ~6.0.0
drizzle-orm ^0.45.1
react ^19.2.4
@ai-sdk/react ^3.0.71
@ai-sdk/openai ^3.0.7
@ai-sdk/google ^3.0.20
@ai-sdk/amazon-bedrock ^4.0.48
@ai-sdk/azure ^3.0.26

1.0.0 - 2025-01-19

Added - First Public Release

Core Features

  • Nexus Chat: Multi-model conversational AI with GPT-5, Claude Opus, and Google Gemini
    • Real-time streaming responses via Server-Sent Events (SSE)
    • Conversation history and organization with folders
    • Model switching mid-conversation
    • Response caching for cost optimization
  • Assistant Architect: No-code AI assistant builder
    • Visual prompt chain designer with variable substitution
    • Knowledge repository integration for context-aware responses
    • Tool integration and function calling
    • Scheduled execution support
  • Knowledge Repositories: Document upload and semantic search
    • Multi-format support (PDF, DOCX, TXT) with OCR via AWS Textract
    • Vector embeddings using pgvector extension
    • Chunk-based search with relevance ranking
    • URL import and processing
  • Model Compare: Side-by-side AI model evaluation
    • Parallel execution across multiple models
    • Token usage and cost analysis
    • Performance metrics (latency, throughput)
    • Export comparison results

Infrastructure

  • AWS ECS Fargate deployment with Application Load Balancer
    • Auto-scaling based on CPU/memory metrics
    • HTTP/2 support for efficient streaming
    • Graviton2 (ARM64) optimization for 20% cost savings
    • Fargate Spot for 70% savings on non-critical workloads
  • Aurora Serverless v2 PostgreSQL database
    • Auto-pause in development (scales to 0 ACU when idle)
    • pgvector extension for similarity search
    • RDS Data API for connection-less queries
    • Automated backups with 7-30 day retention
  • Multi-AZ VPC with comprehensive network topology
    • Public, private-application, private-data, and isolated subnets
    • 14+ VPC endpoints for cost optimization
    • Network ACLs and security groups for defense in depth
    • VPC Flow Logs for security monitoring
  • AWS CDK Infrastructure as Code
    • Reusable constructs for consistent patterns
    • Tag-based IAM least privilege policies
    • Environment-specific configurations (dev, prod)
    • Automated CloudFormation stack deployment

Authentication & Security

  • AWS Cognito with Google OAuth 2.0 integration
  • NextAuth v5 for session management (JWT-based)
  • Role-Based Access Control (RBAC) with tool-level permissions
    • Admin, Teacher, Student roles
    • Granular access to AI models and features
    • Audit logging for sensitive operations
  • Tag-based IAM policies preventing cross-environment access
  • Secrets Manager integration for secure credential storage
  • Parameterized SQL queries preventing injection attacks

Monitoring & Observability

  • Consolidated CloudWatch Dashboard with 115+ widgets
    • Lambda performance metrics
    • ECS health and auto-scaling
    • RDS query performance
    • API latency percentiles
  • AWS Distro for OpenTelemetry (ADOT) for distributed tracing
  • Structured logging with Winston logger
    • Request ID tracking across services
    • PII sanitization via sanitizeForLogging()
    • Configurable log levels per environment
  • Circuit breaker pattern for AI provider reliability
  • Adaptive timeouts based on model capabilities

Documentation

  • 10,000+ lines of architectural diagrams (Mermaid.js)
    • CDK stack dependencies and deployment order
    • VPC network topology with security groups
    • Complete AWS service architecture
    • Database ERD (54 tables across 11 domains)
    • Authentication flow (OAuth 2.0 + Cognito + NextAuth)
    • Request flow diagrams for all major features
    • Assistant Architect execution pipeline
    • Document processing pipeline
    • Streaming architecture (SSE implementation)
  • Comprehensive guides
    • Complete deployment guide with safety checks
    • API reference for all endpoints
    • Error reference with debugging patterns
    • Troubleshooting guide for common issues
    • Security best practices
    • Performance optimization strategies

Cost Optimizations

  • 90% cost reduction vs. individual AI licenses
    • Traditional: $24,000/year (100 users × $20/month ChatGPT Plus)
    • AI Studio: $7,800/year with GPT-5
    • With mixed models (Gemini + GPT-4 mini): ~$2,400/year
  • Lambda PowerTuning: 66% memory reduction for all functions
  • Aurora auto-pause: ~$44/month savings in development
  • S3 Intelligent-Tiering: Automatic cost optimization for documents
  • VPC endpoints: Eliminate NAT gateway data transfer costs

Technical Details

Tech Stack

  • Frontend: Next.js 15 with App Router, React 19, Shadcn UI, Tailwind CSS
  • Backend: ECS Fargate, Aurora Serverless v2 PostgreSQL, AWS Lambda
  • AI Integration: Vercel AI SDK v5 with OpenAI, Anthropic, Google, Bedrock providers
  • Storage: S3 for documents, pgvector for embeddings
  • Authentication: Cognito + NextAuth v5
  • Infrastructure: AWS CDK (TypeScript)
  • Monitoring: CloudWatch + ADOT + OpenTelemetry

Database Schema

  • 54 tables across 11 functional domains
    • Core: users, roles, user_roles, tools, tool_access
    • Nexus Chat: conversations, messages, folders, caching, metrics
    • Assistant Architect: chain_prompts, executions, scheduling, tool_integrations
    • Knowledge: repositories, items, chunks, embeddings (pgvector)
    • MCP Integration: servers, capabilities, audit_logs
    • System: settings, migrations, error_logs

Performance

  • Streaming latency: < 500ms Time to First Token (TTFT)
  • Database queries: < 100ms p95 for most operations
  • API response: < 200ms p95 for non-streaming endpoints
  • Concurrent streams: Supports 100+ simultaneous SSE connections
  • Document processing: Async pipeline with SQS for scalability

Security Features

  • All data processed within district infrastructure (no third-party SaaS)
  • FERPA and COPPA compliance ready
  • Tag-based IAM preventing cross-environment access (dev cannot access prod)
  • Client and server-side input validation
  • Rate limiting and quotas (future enhancement)
  • Security scanning via CodeQL and Dependabot
  • Infrastructure validation via CDK Nag

Migration Notes

This is the first public release. Previously, AI Studio was internal-only to Peninsula School District (PSD401).

Breaking Changes from pre-1.0:

  • Migrated from AWS Amplify to ECS Fargate (different deployment process)
  • Removed legacy Drizzle ORM in favor of RDS Data API
  • Consolidated from 2 VPCs to 1 shared VPC
  • Updated to NextAuth v5 (breaking changes from v4)
  • Upgraded to Next.js 15 (App Router required)