- Added a step to set up QEMU for multi-platform builds
- Specified target platforms for Docker build: linux/amd64 and linux/arm64.
- Corrected the COPY command in Dockerfile to use relative paths correctly.
- changed PostgreSQL and Redis passwords
- replaced Redis service with Valkey
- added Mailpit for local SMTP testing, and update MinIO image version
Updated email service to remove SMTP user and password checks.
* Switch from express to fastify
* Use zod validation for api routes
* Fix type in password reset routes
* Use string as data type for bigint because of serialization issues
* Fix use of user id instead of account id for some file operations
- Extract common AI-related utility functions into a new ai-utils.ts module
- Create a metadata handling library with fetchMetadataForContextItems
- Update chunking and embedding services to use new utility functions
- Improve context retrieval and scoring with recency and hybrid search weights
- Enhance chunk summarization with a more flexible prompt template
- Reorganize types for better type safety and modularity
- Add author name and ID to document and node retrieval services
- Implement bulk author fetching for more efficient metadata retrieval
- Update search result types to include created by information
- Improve context formatting in assistant response generation
- Modify metadata handling to include more detailed author information
- Move assistant response chain and related functions to a dedicated `assistant.ts` library
- Simplify `assistant-response.ts` by delegating complex logic to the new library
- Create type definitions for assistant chain state and related types
- Improve modularity and separation of concerns in AI assistant implementation
- Remove deprecated `ChunkingService` and replace with functional `chunkText` approach
- Simplify response chain graph by removing unnecessary database context steps
- Improve context formatting and citation generation
- Enhance node text extraction for records with field-level details
- Optimize chunking service to handle node metadata more flexibly
- Update LLM service prompts for more precise context generation
