v0.9.0

Release Notes: v0.9.0

Release Date: February 2026
Previous Version: v0.8.1 (December 2, 2025)
Branch: main

---

Summary

Version 0.9.0 is a major platform update delivering significant architectural changes, new features, and infrastructure modernization. With 162 commits, 358 files changed, and 6 merged pull requests, this release represents the largest evolution of the Amplify backend since launch.

Highlights at a Glance

• ⚠️ Breaking Change — All shared environment variables migrated to AWS Parameter Store
• ⚠️ Basic-Ops Elimination — amplify-lambda-basic-ops service removed; existing deployments must migrate
• 🚀 21 New AI Models — GPT-5 series, Claude 4.x, Gemini 2.x, Amazon Nova, Mistral Pixtral
• ⚡ API Gateway Streaming — Migrated from Function URLs to API Gateway Lambda response streaming
• 🔍 Web Search — Admin-configurable web search integration
• 🔔 Critical Error Tracking — Centralized error monitoring with email notifications
• 📊 Configurable Embedding Dimensions — Support for Nova Multimodal Embeddings and flexible vector sizes

---

⚠️ Breaking Changes & Migration

AWS Parameter Store Migration

All shared environment variables have been migrated to AWS Parameter Store. Existing deployments must run the population script before deploying any services:

python3 scripts/populate_parameter_store.py --stage <stage> --dep-name <dep-name>

Services will fail to deploy without this step. See the Migration Guide for complete instructions.

Basic-Ops Service Elimination

The amplify-lambda-basic-ops service has been removed. Its /user-data functionality is now handled by amplify-lambda. If you have the amplify-lambda-basic-ops CloudFormation stack deployed, you must:

1. Check if the stack exists: aws cloudformation describe-stacks --stack-name amplify-<dep-name>-basic-ops-<stage>
2. If it exists, check for user storage data and handle the /user-data API Gateway endpoint conflict
3. Follow Migration Guide Step 3a for detailed instructions

Coordinated Deployment Required

The backend and frontend must be deployed together — deploy the backend first, then rebuild and deploy the frontend container.

---

🆕 What's New

Model Updates

21 new models added, 1 updated, 3 removed (net +18):

Amazon Bedrock:

• amazon.nova-2-multimodal-embeddings-v1:0 — Nova Multimodal Embeddings
• amazon.nova-lite-v1:0, amazon.nova-micro-v1:0, amazon.nova-pro-v1:0 — Nova family

Google Gemini:

• gemini-2.0-flash — Second-generation workhorse (1M context)
• gemini-2.5-flash — Best price-performance with thinking features (1M context, 65K output)
• gemini-2.5-pro — Promoted from preview to stable release

OpenAI GPT:

• gpt-4.1-mini — Latest efficient model
• gpt-5, gpt-5-mini, gpt-5.1, gpt-5.2 — Frontier reasoning and coding models

OpenAI Reasoning:

• o3 — Advanced reasoning
• o4-mini — Cost-effective compact reasoning

Anthropic Claude (Bedrock):

• us.anthropic.claude-3-5-haiku-20241022-v1:0 — Claude 3.5 Haiku
• us.anthropic.claude-opus-4-20250514-v1:0 — Claude 4 Opus
• us.anthropic.claude-opus-4-5-20251101-v1:0 — Claude Opus 4.5
• us.anthropic.claude-sonnet-4-20250514-v1:0 — Claude 4 Sonnet

Mistral:

• us.mistral.pixtral-large-2502-v1:0 — Pixtral Large (124B multimodal)

Removed: amazon.titan-embed-text-v1, text-embedding-3-large, text-embedding-3-small (replaced by Nova embeddings)

Pull Request: #275

---

API Gateway Lambda Response Streaming

Migrated from Function URLs to API Gateway Lambda response streaming:

• API Gateway streaming format with metadata JSON and 8-byte delimiter
• Changed integration type from AWS to AWS_PROXY for streaming support
• Updated Lambda invocation URI to use /2021-11-15/response-streaming-invocations endpoint

Pull Request: #275

---

Web Search Functionality

Admin-configurable web search integration:

• Feature flag support for enabling/disabling web search
• Migrated web search admin configuration to unified admin service
• New webSearch.js module for search operations
• Frontend preferences handling in assistants
• Fixed web_search_preview tool triggering with images
• Integration with OpenAI /v1/responses endpoint

Pull Request: #275

---

Critical Error Tracking & Monitoring

Centralized error monitoring infrastructure:

• Critical error tracker service with database schema
• Critical error processor for event handling
• Email notification system (critical_error_notifier.py)
• Integration across multiple Lambda services
• New criticalLogger.js for JavaScript services
• Schema validation and permissions for critical_errors endpoint

Pull Request: #275

---

Additional Charges & Billing

• Support for additional charges tracking and billing
• Custom cost attribution
• Enhanced billing capabilities

Pull Request: #275

---

🔧 Embeddings & Vector Database

Nova Multimodal Embeddings Support

• Amazon Bedrock Nova Multimodal Embeddings model support
• Updated embedding code for Nova MME compatibility
• Comprehensive documentation (EMBEDDING_DIMENSIONS.md, MODEL_CONFIGURATION.md)

Pull Requests: #264, #268

Configurable Embedding Dimensions

• EMBEDDING_DIM environment variable (defaults to 1536)
• Updated create_table.py for configurable vector dimensions
• populate_parameter_store script updated with EMBEDDING_DIM format
• Backward compatible: existing deployments unaffected (CREATE TABLE IF NOT EXISTS)
• Cross-provider compatibility (Amazon Nova, Cohere Embed v3, OpenAI)

Pull Request: #264

Embedding Improvements

• Enhanced embedding-dual-retrieval with better query handling
• Dead Letter Queue (DLQ) processing for failed embedding chunks (embedding-dlq-handler.py)
• Improved Excel file handling for corrupted files
• Better error handling in visual transcription
• Optimized database interactions
• Enhanced logging for embedding diagnostics

Pull Requests: #262, #275

---

🔐 OAuth & Integrations

Microsoft Azure Admin Consent

• New admin setting for Microsoft Azure integrations
• Store consent settings in DynamoDB alongside existing integrations data
• Check consent setting when user connects an integration

Pull Request: #266

OAuth Enhancements

• Dynamic redirect URI support
• Improved error handling and retry logic for consent errors
• Better origin detection from event headers
• Calendar time zone bug fixes

Pull Request: #275

---

💰 Billing & Cost Tracking

Usage Tracking

• New centralized usageTracking.js module
• Fixed null check for CachedCostPerThousandTokens in recordUsage
• Support for OpenAI cached tokens in usage tracking
• Handle response.completed usage format
• New accounting.js for shared accounting logic

Pull Request: #265

Model Rate Table Updates

• Added pricing for 21 new models
• Updated model_rate_values.csv
• Fixed Bedrock cost tracking bugs
• Improved MTD cost calculations

Pull Requests: #265, #275

---

🐛 Bug Fixes

• Bedrock validation errors: Convert tool roles to user, handle null toolUseId, disable reasoning with tools when incompatible (#265)
• OpenAI usage format: Handle response.completed usage format and cached tokens in openaiUsageTransform (#265)
• Visual transcription: Improved error handling and model retrieval logic (#262)
• IAM policy deployment ordering: Fixed race condition with hardcoded ARN → !Ref for implicit CloudFormation dependencies
• Lambda size limits: Added slimPatterns to strip 85MB Google API discovery cache from assistants-api layer

---

🤖 Assistants & Agent Framework

• Refactored assistant API serverless configuration to use Python requirements layer
• Enhanced user-defined assistants with better state management
• Improved code interpreter integration
• Added OpenAI provider support to agent framework
• Enhanced agent prompt handling and core logic
• Improved scheduled tasks processing
• Removed legacy Docker build files and deployment scripts

Pull Request: #275

---

⚙️ Infrastructure & DevOps

Parameter Store Integration

• All shared environment variables resolved from AWS SSM Parameter Store
• populate_parameter_store.py script for initial population
• Comprehensive migration guide (MIGRATION_README.md)

Var Template Updates

• Updated <env>-var.yml.template with new embedding variables and comments
• All dimensions remain default 1536 unless overridden

Pull Request: #268

Standalone Service Deployment

The following services are deployed independently (not included in serverless-compose.yml):

• amplify-agent-loop-lambda/ — Agent Loop
• amplify-lambda-assistants-api/ — Assistants API
• amplify-lambda-assistants-api-office365/ — Office 365 Integration
• amplify-lambda-assistants-api-google/ — Google Workspace Integration

---

📈 Stats

Metric	Count
Commits	162
Files changed	358
Insertions	+76,919
Deletions	-49,986
Pull Requests	6
New AI Models	21

---

👥 Contributors

• Karely Rodriguez (@karelyrodri) — Lead developer, core infrastructure, JIT provisioning, streaming
• Allen Karns (@karnsab) — Release management, deployment fixes, parameter store migration
• Jason Bradley (@jasonbrd) — Embedding dimensions, Nova MME, model updates, var template
• Seviert (@seviert23) — Microsoft Azure admin consent integration

---

Deployment Instructions

1. Run populate_parameter_store.py (required for all deployments)
2. Handle basic-ops migration if applicable (see Migration Guide Step 3a)
3. Deploy core services: `serverles...

v0.8.1

Release Notes: v0.8.1 to Current Main Branch

Release Date: December 2, 2025
Branch: main

---

Summary

This update to the Amplify GenAI Backend delivers significant improvements across integrations, performance, cost management, and developer experience. With 270 commits and 26 merged pull requests, this release represents a substantial evolution of the platform.

Highlights at a Glance

• SharePoint Integration - Full file browsing and access capabilities
• OpenAI Agent Support - Expanded LLM provider options for agents
• Critical Bug Fixes - Bedrock cost tracking, image processing, RAG deletion
• Performance Optimizations - 40-60% Lambda layer size reduction

---

What's New

SharePoint Integration

Full Microsoft SharePoint integration with advanced file management capabilities.

• Browse SharePoint files and folders with native navigation
• Graph API pagination handling for large document libraries
• Proper mimeType formatting for all file types
• Asynchronous list group operations for improved performance
• Seamless integration with existing Office 365 authentication

Pull Requests: #259, #260

---

OpenAI Provider for Agent Framework

Expand agent capabilities with OpenAI as an LLM provider option.

Benefits:

• More model choices for specialized tasks
• Cost optimization through provider selection
• Fallback options for high availability

Pull Request: #257

---

Critical Cost Tracking Fixes

Resolve critical issues with AWS Bedrock cost calculation and tracking.

Fixed Issues:

• Incorrect token counting for cached inputs
• Missing cost data for certain model types
• Input/output cached token cost handling
• Model rate schema updates

Improvements:

• Enhanced usage cost tracking
• Enhanced billing for multi-account users
• Monthly cost aggregation with detailed history

Pull Request: #258

---

Document Processing

• Prevent Unintended RAG Deletion - Safeguards against accidental data loss
• Daily Cleanup Cron - Automated maintenance for missed RAG secrets
• Improved Lifecycle Management - Better tracking of RAG resources

Pull Request: #254

Image Processing

• OpenAI Image Format Fix - Correct input_text/input_image handling
• Image URL Structure - Use flat string format instead of nested object
• Web Search Prevention - Don't trigger search on image-only queries

Pull Request: #252

Calendar Integration

• Time Zone Handling - Fix critical timezone bug in Google Calendar integration
• Date/Time Accuracy - Improved parsing and formatting

Pull Request: #253

Office 365 Excel Integration

• Worksheet Consistency - Use worksheet_name instead of worksheet_id
• Function Parameter Alignment - Standardize across all Excel operations

Pull Request: #238

---

Performance Improvements

Lambda Layer Optimization

• Size Reduction: 117MB → 60-70MB (40-60% smaller)
• Methods:
  • Strip debug symbols from .so files (15-25MB saved)
  • Remove tests and documentation (15-30MB saved)
  • Eliminate build tools (10-15MB saved)
  • Clean up type stubs and caches

Impact:

• Faster cold starts
• Lower memory usage
• Resolve 262MB deployment limit
• Reduced deployment time

Pagination & Query Optimization

• User Cost Data - Increase limits for large datasets
• DynamoDB Batching - Optimize query patterns
• Multi-Account Support - Efficiently handle 9+ accounts per user

---

Security Enhancements

OAuth Integration

• Dynamic Redirect URIs - Support multiple origins
• Origin Detection - Automatic detection from headers
• Enhanced Error Handling - Retry logic for consent failures
• Secure Serialization - Proper credential handling

Permission Updates

• Standalone Assistant Permissions - Fine-grained data source access
• Drive Integration - Secure file access controls
• DynamoDB Scan Rights - Controlled group table access
• SQS Permissions - Conversation analysis queue access

API Key Management

• amp- Prefix Validation - Consistent key format
• Rate Limiting - Per-key usage controls
• Access Type Controls - Granular permission system
• API Key Migration - Safe ID transition support

---

Cost & Billing

GET /billing/history        # Monthly cost history
                           # Multi-account aggregation
                           # Trend analysis

File Operations

POST /files/query          # Dynamic filtering
                          # Advanced search
GET /files/status          # Embedding status
                          # Async progress tracking

Website Tools

POST /extract-sitemap      # Extract URLs from sitemap
                          # No scraping required
                          # Support maxPages limit

Schema Enhancements

• Database connection validation
• OAuth redirect URI support
• Assistant data source schemas
• Cost history query schemas
• File filter specifications

---

Update required variables:

# New in this release
CHANGE_SET_BOOLEAN: true          # CloudFormation control
LOG_LEVEL: info                   # Logging verbosity
AGENT_STATE_DYNAMODB_TABLE: ...   # Agent state table

Contributors

Special thanks to the following contributors for this release:

• Karely Rodriguez (@karely) - Lead developer, 150+ commits
• Allen Karns (@allenkarns) - JavaScript refactor, layer optimization
• Jason Bradley (@jasonbrd) - Model updates and templates
• Max Moundas (@maxmoundas) - Email webhooks, OAuth integration
• Sam Hays (@samhays) - ID migration system
• Andrew Walker (@FortyAU-Amplify-Team) - Configuration improvements
• Seviert (@seviert23) - Office 365 integrations

---

For detailed technical information, see CHANGELOG.md

Last Updated: December 2, 2025
Version: Unreleased (270 commits ahead of v0.8.0)

Release of v0.8.0

Deployment Notes

• Node version was updated to NodeJS v22 in serverless.yml. This update will require you to run an npm install in the /amplify-lambda-js directory prior to installation.
• Support for markitdown was added to the RAG pipeline. Prior to deploying the /amplify-lambda service navigate to /amplify-lambda/markitdown and run the the markitdown.sh script to create the needed lambda layer.
• Introduction of /pycommon as a package to handle auth and common functionality across all services.

What's Changed

• Advanced Rag - Document Caching by @karelyrodri in #109
• Fix conversation filtering logic to handle cases with missing message… by @karelyrodri in #130
• Refactor group assistant selection logic to refer to DynamoDB query f… by @karelyrodri in #132
• Refactor agent interaction logic and enhance tool handling by @karelyrodri in #143
• Merge samhays-dev into Dev by @gsamuelhays in #144
• Updates to amplify-lambda and assistants by @karelyrodri in #145
• Dev by @jasonbrd in #134
• Refactor access type management and update group permissions logic by @karelyrodri in #151
• Refactor/pycommon by @karelyrodri in #157
• Implement API tools operations, Drive file support, Node 22 upgrade, and various bug fixes by @karelyrodri in #170
• Add websiteUrls feature flag to feature_flags.py by @karelyrodri in #171
• Support OpenAI API Key use in js for models with providers set as OpenAI by @karelyrodri in #172
• Bug fix - Update required fields in register_ops_schema.py by @karelyrodri in #173
• Migrate assistants API services by @karnsab in #179
• Refactor embedding processing to support asynchronous operations by @karelyrodri in #180
• Publish Lambda Loop, Assistants API by @karnsab in #181
• Enhance data source handling and error messaging in group permissions by @karelyrodri in #182
• Bug Fix - Group Assistant by @karnsab in #183
• Bug Fix - Website URL handling and validation in scraping processes by @karelyrodri in #184
• PR Notes: Merge Dev into Main by @karnsab in #185
• Fix error handling in reset_embedding_status_to_starting function by … by @karelyrodri in #186
• Fix error handling in reset_embedding_status_to_starting function by raising exceptions instead of logging them. by @karnsab in #187
• Enhance embedding processing with improved error handling and retry l… by @karelyrodri in #188
• Enhance embedding processing with improved error handling and retry logic by @karnsab in #189
• Add support for GPT-5 by @karnsab in #190
• Add support for GPT-5 by @karnsab in #191
• Bug - Fix Refactor user-defined assistants and conversation analysis … by @karelyrodri in #192
• Update API key rotation logic to include purpose check by @karelyrodri in #194
• Merge dev into main: API key rotation enhancement and account object refactoring by @karnsab in #195
• Updated Amplify rate limit and cost calculations by @karelyrodri in #197
• Integrations - Add data type fixing functionality by @karelyrodri in #199
• Support Thinking Messages + Refactor Get Agent State by @karelyrodri in #200
• Enhancements - Bedrock guardrail, Openai Responses API support, OpenAI web_search tool use, Office365 improvements by @karelyrodri in #203
• Merge Dev Branch into Production by @karnsab in #204
• Bug fix - TokenV1 key generation to use SHAKE256 by @karelyrodri in #205
• API Key Version 1 Hash Bugfix by @karnsab in #206

New Contributors

• @gsamuelhays made their first contribution in #144

Full Changelog: v0.7.5...v0.8.0

Release of v0.7.5

##Notes for Deployment
This release includes improved support for extracting text via Markitdown in the RAG pipeline. You will need to navigate to the ../amplify-lambda/markitdown directory and run ./markitdown.sh prior to deploying the amplify-lambda service.

What's Changed

• Amplify v0.7.5 by @karnsab in #104

• remove duplicate event trigger by @karnsab in #92

• patch event trigger by @karnsab in #93

• Main v0.7 patch by @karnsab in #94

• object-access/serverless.yml missing a Environment Variable and Lambda Permission by @AndDenny16 in #98

• add gemini support by @nahuelpereira-FortyAU in #99

• Add Gemini Support and Backend Enhancements by @karelyrodri in #100

• Recovered updates by @karelyrodri in #101

• Backend Public Update to Dev by @karnsab in #102

• Adding Infrastructure by @karnsab in #103

• Update Dev-Var Example with new vars by @karnsab in #105

• Feature/model updates by @karnsab in #108

New Contributors

• @AndDenny16 made their first contribution in #98
• @nahuelpereira-FortyAU made their first contribution in #99

Full Changelog: v0.7.0...v0.7.5

Release of v0.7.0

Comprehensive Change Log

1. New Features & Integrations

Conversation Registration & Cloud Storage

• New endpoints for registering conversations with validation and improved error handling.
• Added external conversation storage features, including user-level and group-level filtering.

Code Interpreter

• Introduced advanced code interpreter support with logging, error handling, and image support.
• Allowed streaming of code execution events back to the client.

Workflow API

• Added an endpoint to manage LLM workflows with the ability to execute them asynchronously.
• Introduced templates for prompt-based workflows.

Admin UI

• Addition of Admin UI to manage Models, Groups, Variables, Endpoints, etc.
• Addded support for Feature Flags group access to control access to features.
• Updated the admin UI to manage user assistants and group memberships.

---

2. Enhancements & Improvements

Bedrock & Model Providers

• Added Bedrock converse api client with improved tracing capabilities.
• Provided multiple fallback options for cheaper model usage (O1, O3, etc.).
• Improved embedding logic with “embedding provider enums” for flexible usage.

Admin UI & Validator Updates

• Updated the admin UI to manage user assistants and group memberships.
• Enhanced validation on admin endpoints, including group membership checks.
• Improved logging and consolidated admin validations for clarity and security.

Operation Instructions & Execution

• Enhanced the presentation of custom HTTP operations and API documentation.
• Added more robust instructions for operations in the assistant’s flow.
• Moved to parameter-based schemas for vops to ensure consistent request handling.

User Data Handling & Storage

• Added user storage endpoints for a better user experience.
• Introduced caching and token-based user conversation flow.
• Implemented improved username extraction logic and prefix handling.

Feature Flags

• Introduced or expanded feature flags to control oauth integrations, model access, conversation storage, etc.
• Cleaned up code to remove outdated or in-progress flags.

Infrastructure & Deployment

• Refined serverless YAMLs for better environment variable management (e.g., DEP_REGION, hashing table references, etc.).
• Consolidated multiple merges from dev to keep environment variables and function references in sync.
• Enabled Python requirements layering in various Lambdas to reduce cold start times and improve performance.

Logging & Tracing

• Added more detailed logging around key flows (admin ops, conversation ops, code interpreter).
• Implemented additional debugging outputs for error cases, user validations, and model usage.

---

3. Bug Fixes

Memory & DynamoDB

• Addressed a variety of syntax and table name issues that caused read/write failures.
• Fixed a bug where memory feature flags did not properly initialize.

Admin UI

• Resolved issues with the admin panel failing to load group assistants or incorrectly validating user membership.

Conversation Handling

• Fixed a bug causing empty remote conversations to break UI flows.
• Corrected invalid conversation statuses in certain edge cases.
• Stopped repeated circular dependencies in chat endpoints.

Code Interpreter

• Fixed image read/write logic that caused client crashes.
• Improved error messages for unhandled or malformed code blocks.

Environment Variables & Infrastructure

• Corrected references to environment variables (e.g., HASH_FILES_DYNAMO_TABLE, DEP_REGION) in serverless configs.
• Removed Anthropic and Mistral references in code to avoid build breaks in the public branch.

Model & Embedding

• Patched Bedrock embedding requests that sent incomplete data.
• Fixed the final response format for O1 and O3 models to match the required schema.

---

4. Refactoring & Cleanup

Codebase Refactoring

• Moved Oauth logic to separate modules for better maintainability.
• Transitioned older doc references and test files out of the public branch.

Documentation & Readmes

• Added or updated READMEs for new integrations.
• Updated operation docs for custom HTTP ops, memory usage, and new endpoints.

Removal of Deprecated/Unused Services

• Removed references to external API services not published in the public branch.
• Purged test stubs, old environment variables, and data source references.

---

Release of v.0.6.0

What's Changed

• Feature Release by @karnsab in #26
• Add Features for version v0.6.0 by @karnsab in #28
• Features:

• Remote Conversations
• Prompt Highlighter
• Artifacts
• Settings Menu
• Conversation Forking
• Group Assistants
• API
• Add support for Bedrock Inference Endpoints

Full Changelog: v0.5.0...v0.6.0

Initial Release of Amplify GenAI Backend

What's Changed

• update service name by @karnsab in #2
• clean up requirements.txt by @karnsab in #3
• clean up duplicates in requirements.txt by @karnsab in #4
• add bedrock to lambda js iam by @karnsab in #5
• add rquirements back by @karnsab in #6
• open ai api support by @karnsab in #7
• gpt 4o and sonnet 3.5 support by @karnsab in #8
• Clean up deployment by @karnsab in #9
• update concurrency for index_market_item by @karnsab in #10
• Add MIT License by @karnsab in #12
• fix environment for get_emails by @karnsab in #13
• Add powerpoint templates to misc deployment files by @karnsab in #14
• Update object access/embedding to python 3.11. ad sonnet 3.5 to most … by @karnsab in #16
• Update object access/embedding to python 3.11. ad sonnet 3.5 to most... by @karnsab in #17
• Fix object access serverless compose entry by @karnsab in #18
• fix object access path in serverless compose... by @karnsab in #19
• Add support for Bedrock Embeddings by @karnsab in #20
• Re-ordered variables and groupings in sample template; adjust docs by @jasonbrd in #21

New Contributors

• @karnsab made their first contribution in #2
• @jasonbrd made their first contribution in #21

Full Changelog: https://github.com/gaiin-platform/amplify-genai-backend/commits/v0.5.0