Release notes
1.1.3 (February 27, 2026)
Improvements:
- Support including files in YAML using
module:filesyntax
1.1.2 (February 27, 2026)
Bug fixes and improvements:
- Fixed parsing of nested curly braces in string expressions
- Changed default console logging level to WARN for cleaner output
- Minor warning fixes
1.1.1 (February 25, 2026)
Bug fixes and improvements:
- Improved error handling in custom tools to properly propagate exceptions (#110)
- Fixed issue where exceptions from failed evaluations were wrapped into string messages too early
- Ensured error conversion only occurs at the boundary between tool code and LLM interaction
- Fixed pricing information for Opus 4.5 model
- Fixed logging for outgoing/incoming LLM messages
New features:
- Added user setting to control streaming in chat interface
- Added CLI arguments to environment variables via
env.CLI_ARGS
Improvements:
- Changed Pydantic model serialization to use
exclude_none=Truefor cleaner output - Added extra stderr output during evaluation runs for better debugging
1.1.0 (February 22, 2026)
New features:
- Added
llm-workers-evaluationpackage with evaluation framework for testing LLM scripts (#98)- New
llm-workers-evaluateCLI tool for running evaluation suites - YAML-based evaluation suite configuration
- Score calculation with automatic averaging across iterations, tests, and suites (#101)
- Data and tool merging from shared → suite → test levels
- Logging support in evaluation framework (#102)
- Token usage reporting in evaluation results (#103)
- Confidence interval reporting in evaluation results (#105)
- New
- Added parallel execution support to
for_eachstatement (#104) - Simplified built-in tool usage (#99)
Improvements:
- Comprehensive logging overhaul with scoped loggers and improved formatting (#107, #108, #109)
- Added support for Pydantic models in
print_json - Returned type conversion functions (
list,tuple,dict) to Starlark environment
Bug fixes:
- Fixed exception handling in parallel
for_each - Fixed formatting of multiline strings
- Fixed input data wrapping in
evalstatements
1.0.0 (January 20, 2026)
- Misc documentation updates
- Allow non-ASCII characters in JSON dumps
- Final release following code freeze
1.0.0-rc9 (January 18, 2026)
Tool improvements and scripting enhancements:
- Improved
read_filetool with better defaults and options (#89) - Enhanced UI hints for FilesystemToolkit tools (#90)
- Simplified
grep_filesparameters (#91) - Added
for_eachstatement for iterating in custom tools (#93) - Exposed shared tools to Starlark scripts (#94)
- Improved CLI support with better tool call output (#95)
- Added UI hint support for
callstatement (#96)
1.0.0-rc8 (January 9, 2026)
Windows compatibility and release automation improvements:
- Fixed Windows compatibility by making
pwdandgrpmodules optional imports in filesystem tools (#88) - Added automated release scripts for version management, building, and publishing
- Improved release process documentation
1.0.0-rc7 (December 30, 2025)
Last substantial changes before 1.0.0 code freeze, focusing on modularization and session management:
- Split project into sub-modules:
llm-workers(core),llm-workers-console(chat UI), andllm-workers-tools(CLI) - Added support for saving and resuming chat sessions
- Introduced
!includeand!requireYAML tags for modular configuration - Improved handling of multiple async MCP sessions
- Streamlined project dependencies
1.0.0-rc6
Accidentally skipped version.
1.0.0-rc5 (December 26, 2025)
Core infrastructure improvements and model support expansion:
- Completed basic agent harness implementation
- Refactored tools importing for better extensibility
- Fixed CLI functionality issues
- Refactored statements set for cleaner architecture
- Improved variable referencing system
- Enhanced custom tool functionality with additional features
- Reorganized script structure
- Implemented comprehensive token usage tracking
- Refreshed and expanded model support (including Google models)
- Updated documentation
1.0.0-rc4 (December 5, 2025)
UI polish and configuration improvements:
- Added
${env.}substitution in environment descriptions - Miscellaneous UI fixes
1.0.0-rc3 (December 5, 2025)
Tool functionality improvements and model updates:
- Fixed tool registration issues in
build_llm_tool - Fixed broken nested tool invocation UI hints
- Upgraded to Claude Sonnet 4.5 as default model
1.0.0-rc2 (December 3, 2025)
UI enhancements for streaming and tool parameters:
- Fixed markdown streaming functionality
- Added selective display of tool call arguments in MCP tool UI hints
1.0.0-rc1 (December 2, 2025)
Major infrastructure improvements and MCP integration:
- Replaced callbacks with notifications via astream
- Added basic MCP (Model Context Protocol) support
- Improved environment variable configuration
- Fixed LLM tool to return plain output
- Added
/newcommand support
0.1.0-rc13 (October 4, 2025)
Comprehensive UX improvements:
- Added “Thinking…” prompt during LLM requests
- Improved commands support
- Enhanced token usage and cost reporting
- Moved UX-related switches to user config
- Changed confirmation to normal line input
- Improved handling of missing values from .env files
- Fixed reasoning display for OpenAI models
- Added @-triggered filename completion
- Passed token usage from LLM-backed tools
- Reduced colorfulness of chat interface
0.1.0-rc12 (September 20, 2025)
Chat interface and system enhancements:
- Added ability to switch models in chat
- Kept run Python scripts for audit purposes
- Improved caching
- Added
/export <name>command
0.1.0-rc11 (September 11, 2025)
Model management improvements:
- Simplified model management functionality
0.1.0-rc10 (August 20, 2025)
Configuration flexibility enhancements:
- Added support for resolving using dynamic keys
- Implemented inline tool definition
- Added support for inline tool and model configuration
0.1.0-rc9 (July 1, 2025)
Tool result handling improvements:
- Added LLM tool support to filter results to JSON only
- Improved handling of tools returning non-string results
0.1.0-rc8 (July 1, 2025)
Chat output enhancements:
- Added experimental markdown output support to chat functionality
0.1.0-rc7 (June 19, 2025)
LLM tool customization:
- Added ability to specify system prompt as LLM tool parameter
0.1.0-rc6 (June 19, 2025)
Template and script organization:
- Added support for references to nested elements in templates
- Introduced “shared” section in LLM scripts
0.1.0-rc5 (June 11, 2025)
Bug fixes:
- Fixed RunPythonScriptTool to use configured Python executable instead of system default
0.1.0-rc4 (June 11, 2025)
Type system and UI enhancements:
- Added support for missing intermediate tool UI hints
- Added support for generic lists and dictionaries as parameter types
0.1.0-rc3 (June 11, 2025)
User interface and interaction improvements:
- Added support for hiding UI hints
- Implemented single-key confirmations
- Added support for showing nested tool executions
- Introduced approval tools
0.1.0-rc2 (June 10, 2025)
User experience enhancements:
- Added User Input tool
- Added optional welcome banner for users
0.1.0-rc1 (May 30, 2025)
Stability and robustness improvements:
- Improved error handling in main chat session loop
- Completed initial documentation
- Reconsidered and improved logging
- Enhanced tool error handling
- Fixed caching issues
- Added throttling configuration for AWS Bedrock models
- Added support for literal types in custom tools
0.1.0-alpha5 (March 26, 2025)
Initial alpha release with core features:
- Added Claude 3.7 thinking mode support via AWS Bedrock
- Implemented loading YAML files from resources
- Added environment-specific configuration
- Simplified tool run confirmations
- Improved handling of
return_directtools - Added auto-open for new/updated files in current working directory