Product Updates

​

Highlights

• Introduced pxtf.map() as a replacement for the >> operator to create JsonMappers
• Added string concatenation operations (+ and *) support in arithmetic expressions
• Incorporated import operations into table_create and insert methods
• Added access method for embedding indices
• Switched to pixeltable-yolox for computer vision functionality

​

Enhancements

Restructured documentation for improved navigation and clarity Added table IDs to Pixeltable markdown documentation Added create_parents option to create_dir to automatically create missing parent directories Improved JsonMapper functionality with new unit tests for QueryTemplateFunction

​

Fixes

Fixed event loop debug logging Resolved syntax errors in documentation Addressed bugs in directory operations, particularly when drop_dir() is the first operation after catalog load Fixed issues with chained tool calling Corrected bug involving @pxt.query instances with default parameter values Improved JsonPath serialization

​

Highlights

• Introduced linting for improved code quality
• Added just-in-time initialization for spaCy, improving pxt.init() performance
• Made catalog changes to prepare for concurrency support

​

Enhancements

• Added video index to cookbook
• Updated configurations page to match API reference
• Added MCP to documentation
• Improved documentation with updated vision search examples

​

Fixes

• Implemented graceful failure handling for backwards incompatibility in computed column UDF calls
• Various bugfixes and improvements
• Updated Label Studio job to Python 3.10 in nightly CI

​

Highlights

• Enhanced OpenAI/Anthropic integration with support for multiple invocations of the same tool in tool calling logic

​

Highlights

• Added Deepseek integration
• Implemented data sharing logic for publishing snapshots
• Enhanced UDF handling in computed columns
• Migrated to Mintlify documentation

​

Enhancements

• Improved test suite with pytest fixtures for Hugging Face embedding models
• Enabled view creation from dataframes with select clause
• Updated PyAV to 14.2 and WhisperX to 3.3.1
• Improved handling of relative pathnames and filenames with unusual characters

​

Documentation

• Fixed documentation for stored attribute on computed columns
• Added audio file examples

​

Development & Infrastructure

• Updated llama_cpp version (disabled in non-Linux CI)
• Implemented release script fixes for Poetry 2.0

​

Highlights

• Added support for OpenAI reasoning models
• Introduced tables as UDFs for more modular workflows
• Implemented AudioSplitter support for audio processing
• Enabled all types of computed columns to be unstored for flexibility
• Added support for variable parameters in query limit() clause
• Enhanced data management with a packager for table data
• Updated PostgreSQL to version 16.8 and pgvector to 0.8.0

​

Enhancements

• Improved parallel execution capabilities
• Added support for generalized arrays (unparameterized/with only a dtype)
• Allowed numpy.ndarray arrays to be used as Literal constants
• Enhanced type checking for tests package
• Improved handling of collections with all constants as Literals
• Converted more UDFs to async for better performance
• Added verbose system config option for improved debugging

​

Fixes

• Fixed FastAPI integration bug
• Resolved issues with AsyncConnectionPool behavior
• Improved test resiliency and reliability
• Fixed tiktoken dependency issue
• Corrected validity of column error properties
• Upgraded httpcore for better compatibility
• Fixed notebook test failures

​

Development & Infrastructure

• Added archive functionality for Pixeltable logs from every test run
• Improved CI/CD workflow with tmate in detached mode
• Enhanced documentation with updates to numerous guides
• Streamlined API syntax for better developer experience
• Updated example applications to use new query syntax

​

Highlights

• Enhanced Function Support with multiple signatures capability for Functions, UDFs, and UDAs
• Improved Data Validation with JSON Schema validation for JsonType columns
• Enhanced Database Reliability by changing SQL Engine isolation level to ‘REPEATABLE READ’

​

Enhancements

• Added ifexists parameter to create* APIs for better control
• Improved DataFrame docstrings for better documentation
• Fixed indexed column loading for views
• Enhanced type validation by preventing bool literals in int columns
• Improved handling of index name conflicts

​

Documentation & Examples

• Updated Discord Bot documentation
• Added Gemini integration examples

​

Fixes

• Fixed assertion in ReloadTester
• Resolved pgserver-related issues for Docker and windows setup

​

Highlights

• Added Python 3.13 Support
• Introduced basic joins functionality for tables
• Added Gemini AI integration
• Implemented Parquet export API
• Extended document support to include .txt files

​

Enhancements

• Added test utility for query result verification after catalog reload
• Fixed Optional vs. Required handling in astype()
• Updated Ollama integration for version 0.4.0
• Added graceful error handling when using dropped catalog.Tables
• Reorganized docs and examples folders
• Added feature guide for time zones
• Made Tables, DataFrames, and Expressions repr output more user-friendly

​

Fixes

• Fixed string comparison to use != instead of ‘is not’
• Resolved various development environment configuration issues

​

New Contributors

• @jacobweiss2305 made his first contribution

​

Highlights

• Added Context-Aware Discord Bot with Semantic Search Capabilities
• Introduced TileIterator for efficient data processing
• Migrated to torchaudio from librosa for improved audio preprocessing

​

Enhancements

• Implemented reusable retry script for CI workflows
• Added configuration documentation (config.md)
• Enhanced Function bindings with partial support
• Fixed backwards-incompatible Mistral API changes
• Improved create_insert_plan functionality
• Disabled sentence_transformers tests on linux ARM for better stability
• Updated README.md with clearer organization
• Added support for table/column handles in APIs

​

Highlights

• Added support for Ollama, llama_cpp, and Replicate
• Switched FrameIterator to PyAV and added XML document type support
• Added Voxel51 integration for computer vision workflows
• Implemented custom type hints for all Pixeltable types
• Added support for converting aggregate FunctionCalls to SQL
• Streamlined create_view API and enhanced documentation

​

Development & Infrastructure

• Updated CI/CD configuration and Makefile
• Upgraded GitHub Actions to use macos-13
• Limited ubuntu-arm64 and ubuntu-x64-t4 to scheduled runs
• Added Image.point() to API
• Improved type-checking correctness across packages
• Enhanced documentation and display for new type hint pattern

​

Fixes

• Fixed issues in working-with-huggingface notebook
• Resolved Replicate notebook compatibility with external URLs
• Ensured correct nullability in FunctionCall return types
• Added exception raising during add_column() errors
• Allowed @query redefinition in notebook scope
• Updated BtreeIndex.str_filter implementation

​

Enhancements

• Added support for loading Hugging Face datasets containing images
• Implemented LRU eviction in FileCache for improved memory management
• Enhanced JSON path functionality to allow getitem syntax
• Updated iterators to handle None values as input

​

Fixes

• Resolved an issue with the Together AI image endpoint

​

Enhancements

• Initial support for converting FunctionCalls to SQL
• Added comprehensive time zone handling
• Improved type-checking correctness for catalog, functions, and ext packages
• Introduced integration with Mistral AI and Anthropic
• Added a new tutorial on computed columns

​

Improvements

• Made mistune an optional dependency

​

Fixes

• Resolved a circularity issue in database migration for schema version 19 -> 20

​

Enhancements

• Improved type-checking system with groundwork and performance improvements
• Added cross-links to docstrings
• Enhanced create_table to accept DataFrame directly
• Updated Postgres to version 16.4 and pgvector to 0.7.4
• Implemented Notebook CI and Nightly CI

​

Fixes

• Fixed unit test for Together AI integration
• Resolved notebook regressions
• Updated to psycopg3 as Postgres driver
• Cleaned up Table class namespace
• Fixed JSON serialization and literal handling

​

Highlights

• Optimized data loading with StoreBase.load_column()
• Added support for lists, dictionaries, and non-numpy datatypes in import_pandas
• Enhanced video frame extraction control in FrameIterator
• Added UDF draw_bounding_boxes() for object detection visualization
• Migrated to Pixeltable-compatible pgserver fork
• Made all column types stored by default

​

New Features

• Added import_json() and import_rows() functions
• Expanded timestamp functions library
• Added aggregate make_list() function

​

Improvements

• Simplified method call syntax
• Enhanced notebook experience
• Improved test coverage and automation

​

Fixes

• Updated database version
• Removed support for Python datetime.date
• Improved CSV import with nullable types

​

Features

• Added Label Studio integration with pre-signed URLs for S3 buckets
• Enhanced compatibility with newer versions of label-studio-sdk
• Added new String functions
• Introduced new tutorial about Tables and Data Operations

​

Features

• Added force=True option for drop_table and drop_dir functions
• Enhanced API usability and functionality
• Updated tutorials covering OpenAI, Together, Fireworks, and indexing