Release v1.1.4

Changelog

All notable changes to this project will be documented in this file.

[Unreleased]

Added

• Qwen Tool Call Support: Added support for Qwen2.5-coder's XML-based tool calling format
  • New parser function extracts tool calls from <tool_call> XML tags
  • Automatically converts Qwen format to standard tool_calls format
  • Added detection logic prioritizing Qwen format after standard format
  • Enhanced logging for Qwen tool call processing

Fixed

• Qwen2.5-coder Compatibility: Qwen models now properly execute tool calls instead of outputting them as text
  • Resolves issue where tool calls appeared as JSON in conversation responses
  • Tool calls wrapped in XML tags are now parsed and executed correctly

[1.1.0] - 2025-12-29

Fixed

• Gemma3 Think Block Issue: Fixed issue where Gemma3 model responses were showing raw blocks instead of user-facing text
  • Enhanced filter function to handle unclosed tags
  • Added debug logging to track filtering operations
  • Implemented fallback responses when filtering results in empty text
  • System prompt now explicitly instructs models not to use tags

Added

• Smart fallback response generation when model output is filtered to empty
• Automatic action summaries (e.g., "Done! I've turned off light.desk_lamp")
• Enhanced debug logging for response processing pipeline
• Raw response logging before filtering (debug level)

Changed

• System prompt updated with explicit "Response Format" section
• Prohibits use of tags in model instructions
• More natural confirmation examples in system prompt

[1.0.0] - 2024-01-XX

Added

• Initial release with Ollama conversation agent
• Support for light control (on/off, brightness)
• Support for climate control (temperature setting)
• Standard Ollama tool calling format
• Gemma3-tools format compatibility
• Dynamic entity exposure
• Conversation history (10 messages)
• UI-based configuration flow

Release v1.1.3

Changelog

All notable changes to this project will be documented in this file.

[Unreleased]

Added

• Qwen Tool Call Support: Added support for Qwen2.5-coder's XML-based tool calling format
  • New parser function extracts tool calls from <tool_call> XML tags
  • Automatically converts Qwen format to standard tool_calls format
  • Added detection logic prioritizing Qwen format after standard format
  • Enhanced logging for Qwen tool call processing

Fixed

• Qwen2.5-coder Compatibility: Qwen models now properly execute tool calls instead of outputting them as text
  • Resolves issue where tool calls appeared as JSON in conversation responses
  • Tool calls wrapped in XML tags are now parsed and executed correctly

[1.1.0] - 2025-12-29

Fixed

• Gemma3 Think Block Issue: Fixed issue where Gemma3 model responses were showing raw blocks instead of user-facing text
  • Enhanced filter function to handle unclosed tags
  • Added debug logging to track filtering operations
  • Implemented fallback responses when filtering results in empty text
  • System prompt now explicitly instructs models not to use tags

Added

• Smart fallback response generation when model output is filtered to empty
• Automatic action summaries (e.g., "Done! I've turned off light.desk_lamp")
• Enhanced debug logging for response processing pipeline
• Raw response logging before filtering (debug level)

Changed

• System prompt updated with explicit "Response Format" section
• Prohibits use of tags in model instructions
• More natural confirmation examples in system prompt

[1.0.0] - 2024-01-XX

Added

• Initial release with Ollama conversation agent
• Support for light control (on/off, brightness)
• Support for climate control (temperature setting)
• Standard Ollama tool calling format
• Gemma3-tools format compatibility
• Dynamic entity exposure
• Conversation history (10 messages)
• UI-based configuration flow

Release v1.1.2

Changelog

All notable changes to this project will be documented in this file.

[1.1.0] - 2025-12-29

Fixed

• Gemma3 Think Block Issue: Fixed issue where Gemma3 model responses were showing raw <think> blocks instead of user-facing text
  • Enhanced _filter_think_blocks() to handle unclosed <think> tags
  • Added debug logging to track filtering operations
  • Implemented fallback responses when filtering results in empty text
  • System prompt now explicitly instructs models not to use <think> tags

Added

• Smart fallback response generation when model output is filtered to empty
• Automatic action summaries (e.g., "Done! I've turned off light.desk_lamp")
• Enhanced debug logging for response processing pipeline
• Raw response logging before filtering (debug level)

Changed

• System prompt updated with explicit "Response Format" section
• Prohibits use of <think> tags in model instructions
• More natural confirmation examples in system prompt

[1.0.0] - 2024-01-XX

Added

• Initial release with Ollama conversation agent
• Support for light control (on/off, brightness)
• Support for climate control (temperature setting)
• Standard Ollama tool calling format
• Gemma3-tools format compatibility
• Dynamic entity exposure
• Conversation history (10 messages)
• UI-based configuration flow

Release v1.1.1

Changelog

All notable changes to this project will be documented in this file.

[1.1.0] - 2025-12-29

Fixed

• Gemma3 Think Block Issue: Fixed issue where Gemma3 model responses were showing raw <think> blocks instead of user-facing text
  • Enhanced _filter_think_blocks() to handle unclosed <think> tags
  • Added debug logging to track filtering operations
  • Implemented fallback responses when filtering results in empty text
  • System prompt now explicitly instructs models not to use <think> tags

Added

• Smart fallback response generation when model output is filtered to empty
• Automatic action summaries (e.g., "Done! I've turned off light.desk_lamp")
• Enhanced debug logging for response processing pipeline
• Raw response logging before filtering (debug level)

Changed

• System prompt updated with explicit "Response Format" section
• Prohibits use of <think> tags in model instructions
• More natural confirmation examples in system prompt

[1.0.0] - 2024-01-XX

Added

• Initial release with Ollama conversation agent
• Support for light control (on/off, brightness)
• Support for climate control (temperature setting)
• Standard Ollama tool calling format
• Gemma3-tools format compatibility
• Dynamic entity exposure
• Conversation history (10 messages)
• UI-based configuration flow

v1.1.0 - Gemma3 Think Block Fix

Ollama Conversation Agent v1.1.0

🎯 Overview

This release fixes a critical issue where the Gemma3 model was outputting raw <think> blocks instead of user-facing responses when controlling devices.

🐛 Bug Fixes

Gemma3 Think Block Issue

Problem: When using Gemma3, device control responses showed internal reasoning blocks like:

<think> Alright, the user just turned off the desk lamp. Let me make sure I followed all the instructions...

Solution: Implemented comprehensive filtering and fallback mechanisms:

• Enhanced regex patterns to handle both complete and unclosed <think> tags
• Added intelligent fallback response generation when filtering results in empty text
• Updated system prompt to explicitly prohibit <think> tags in model outputs
• Added comprehensive debug logging for troubleshooting

✨ New Features

• Smart Fallback Responses: When model outputs only reasoning, automatically generates natural confirmations (e.g., "Done! I've turned off light.desk_lamp.")
• Enhanced Debug Logging: Track response filtering operations and model behavior
• Improved System Prompt: Clearer instructions for models on response formatting

🔧 Technical Changes

Modified Files

• conversation.py: Enhanced _filter_think_blocks() function with dual regex patterns and fallback logic
• manifest.json: Version bump to 1.1.0
• CHANGELOG.md: Comprehensive changelog with all changes

New Files

• FIX_SUMMARY.md: Technical details of the fix implementation
• TESTING.md: Deployment and testing guide

📦 Installation

Method 1: HACS (Recommended)

1. Open HACS in Home Assistant
2. Go to Integrations
3. Search for "Ollama Conversation"
4. Click Update (if already installed) or Install

Method 2: Manual Installation

1. Download ollama-conversation-v1.1.0.zip
2. Extract to your Home Assistant custom_components directory:

   unzip ollama-conversation-v1.1.0.zip -d /config/custom_components/

3. Restart Home Assistant

Method 3: Direct Copy

cd /path/to/homeassistant-ollama-agent
scp -r custom_components/ollama_conversation your-ha-server:/config/custom_components/

🧪 Testing

After installation, test with these commands:

1. Light Control: "Turn off the desk lamp"
2. Climate Control: "Set the bedroom to 72 degrees"
3. Brightness: "Turn on the kitchen light to 50%"

Expected Results:

• ✅ No <think> tags visible to users
• ✅ Natural, concise confirmations
• ✅ Proper device control execution

🐞 Debug Mode

To enable detailed logging, add to configuration.yaml:

logger:
  default: info
  logs:
    custom_components.ollama_conversation: debug

🔗 Links

• Repository: https://github.com/cronus42/homeassistant-ollama-agent
• Issues: https://github.com/cronus42/homeassistant-ollama-agent/issues
• Documentation: See README.md and INSTALL.md

📋 Full Changelog

See CHANGELOG.md for complete details.

🙏 Credits

Developed by goose (Block's AI agent) as part of the open-source Goose project.

---

Compatibility: Home Assistant 2024.x or later, Ollama with function-calling support
Supported Models: Gemma3, function-gemma, and other Ollama models with tool calling

Release v1.0.6

Changelog

All notable changes to this project will be documented in this file.

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

[1.0.1] - 2025-12-27

Fixed

• Bug #1: Fixed 500 Internal Server Error when conversing with assistant
  • Added better error handling and logging in chat API calls
  • Increased timeout from 30s to 60s for tool-heavy conversations
  • Added detailed debug logging to diagnose API issues
  • Fixed response parsing to handle edge cases
• Bug #2: Fixed "This assistant cannot control your home" warning
  • Added supported_features property returning ConversationEntityFeature.CONTROL
  • Integration now properly declares device control capabilities to Home Assistant

Changed

• Improved error messages with more descriptive information
• Added async_set_speech() call to properly set response text in intent
• Enhanced tool argument parsing to handle both dict and string formats
• Better exception handling with user-friendly error messages

Improved

• More comprehensive logging throughout the integration
• Better JSON parsing error handling
• Response status validation before parsing

[1.0.0] - 2025-12-26

Added

• Initial release of Ollama Conversation integration
• Full Ollama API support with tool/function calling
• UI-based configuration flow
• Support for light control (on/off, brightness)
• Support for climate control (temperature setting)
• Conversation history management (last 10 messages)
• Configurable parameters (temperature, context_window, top_p, top_k)
• Comprehensive error handling
• Model selection dropdown
• Connection validation
• Unit tests and functional tests
• Complete documentation

Features

• Local AI processing via Ollama
• Natural language device control
• Multi-turn conversation support
• Tool execution with result feedback
• Home Assistant service integration
• HACS compatibility

Documentation

• Detailed README with usage examples
• Implementation summary
• Troubleshooting guide
• Architecture documentation
• API endpoint documentation

Release v1.0.5

Changelog

All notable changes to this project will be documented in this file.

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

[1.0.1] - 2025-12-27

Fixed

• Bug #1: Fixed 500 Internal Server Error when conversing with assistant
  • Added better error handling and logging in chat API calls
  • Increased timeout from 30s to 60s for tool-heavy conversations
  • Added detailed debug logging to diagnose API issues
  • Fixed response parsing to handle edge cases
• Bug #2: Fixed "This assistant cannot control your home" warning
  • Added supported_features property returning ConversationEntityFeature.CONTROL
  • Integration now properly declares device control capabilities to Home Assistant

Changed

• Improved error messages with more descriptive information
• Added async_set_speech() call to properly set response text in intent
• Enhanced tool argument parsing to handle both dict and string formats
• Better exception handling with user-friendly error messages

Improved

• More comprehensive logging throughout the integration
• Better JSON parsing error handling
• Response status validation before parsing

[1.0.0] - 2025-12-26

Added

• Initial release of Ollama Conversation integration
• Full Ollama API support with tool/function calling
• UI-based configuration flow
• Support for light control (on/off, brightness)
• Support for climate control (temperature setting)
• Conversation history management (last 10 messages)
• Configurable parameters (temperature, context_window, top_p, top_k)
• Comprehensive error handling
• Model selection dropdown
• Connection validation
• Unit tests and functional tests
• Complete documentation

Features

• Local AI processing via Ollama
• Natural language device control
• Multi-turn conversation support
• Tool execution with result feedback
• Home Assistant service integration
• HACS compatibility

Documentation

• Detailed README with usage examples
• Implementation summary
• Troubleshooting guide
• Architecture documentation
• API endpoint documentation

Release v1.0.4

Changelog

All notable changes to this project will be documented in this file.

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

[1.0.1] - 2025-12-27

Fixed

• Bug #1: Fixed 500 Internal Server Error when conversing with assistant
  • Added better error handling and logging in chat API calls
  • Increased timeout from 30s to 60s for tool-heavy conversations
  • Added detailed debug logging to diagnose API issues
  • Fixed response parsing to handle edge cases
• Bug #2: Fixed "This assistant cannot control your home" warning
  • Added supported_features property returning ConversationEntityFeature.CONTROL
  • Integration now properly declares device control capabilities to Home Assistant

Changed

• Improved error messages with more descriptive information
• Added async_set_speech() call to properly set response text in intent
• Enhanced tool argument parsing to handle both dict and string formats
• Better exception handling with user-friendly error messages

Improved

• More comprehensive logging throughout the integration
• Better JSON parsing error handling
• Response status validation before parsing

[1.0.0] - 2025-12-26

Added

• Initial release of Ollama Conversation integration
• Full Ollama API support with tool/function calling
• UI-based configuration flow
• Support for light control (on/off, brightness)
• Support for climate control (temperature setting)
• Conversation history management (last 10 messages)
• Configurable parameters (temperature, context_window, top_p, top_k)
• Comprehensive error handling
• Model selection dropdown
• Connection validation
• Unit tests and functional tests
• Complete documentation

Features

• Local AI processing via Ollama
• Natural language device control
• Multi-turn conversation support
• Tool execution with result feedback
• Home Assistant service integration
• HACS compatibility

Documentation

• Detailed README with usage examples
• Implementation summary
• Troubleshooting guide
• Architecture documentation
• API endpoint documentation

Release v1.0.3

Changelog

All notable changes to this project will be documented in this file.

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

[1.0.1] - 2025-12-27

Fixed

• Bug #1: Fixed 500 Internal Server Error when conversing with assistant
  • Added better error handling and logging in chat API calls
  • Increased timeout from 30s to 60s for tool-heavy conversations
  • Added detailed debug logging to diagnose API issues
  • Fixed response parsing to handle edge cases
• Bug #2: Fixed "This assistant cannot control your home" warning
  • Added supported_features property returning ConversationEntityFeature.CONTROL
  • Integration now properly declares device control capabilities to Home Assistant

Changed

• Improved error messages with more descriptive information
• Added async_set_speech() call to properly set response text in intent
• Enhanced tool argument parsing to handle both dict and string formats
• Better exception handling with user-friendly error messages

Improved

• More comprehensive logging throughout the integration
• Better JSON parsing error handling
• Response status validation before parsing

[1.0.0] - 2025-12-26

Added

• Initial release of Ollama Conversation integration
• Full Ollama API support with tool/function calling
• UI-based configuration flow
• Support for light control (on/off, brightness)
• Support for climate control (temperature setting)
• Conversation history management (last 10 messages)
• Configurable parameters (temperature, context_window, top_p, top_k)
• Comprehensive error handling
• Model selection dropdown
• Connection validation
• Unit tests and functional tests
• Complete documentation

Features

• Local AI processing via Ollama
• Natural language device control
• Multi-turn conversation support
• Tool execution with result feedback
• Home Assistant service integration
• HACS compatibility

Documentation

• Detailed README with usage examples
• Implementation summary
• Troubleshooting guide
• Architecture documentation
• API endpoint documentation

Release v1.0.2

Changelog

All notable changes to this project will be documented in this file.

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

[1.0.1] - 2025-12-27

Fixed

• Bug #1: Fixed 500 Internal Server Error when conversing with assistant
  • Added better error handling and logging in chat API calls
  • Increased timeout from 30s to 60s for tool-heavy conversations
  • Added detailed debug logging to diagnose API issues
  • Fixed response parsing to handle edge cases
• Bug #2: Fixed "This assistant cannot control your home" warning
  • Added supported_features property returning ConversationEntityFeature.CONTROL
  • Integration now properly declares device control capabilities to Home Assistant

Changed

• Improved error messages with more descriptive information
• Added async_set_speech() call to properly set response text in intent
• Enhanced tool argument parsing to handle both dict and string formats
• Better exception handling with user-friendly error messages

Improved

• More comprehensive logging throughout the integration
• Better JSON parsing error handling
• Response status validation before parsing

[1.0.0] - 2025-12-26

Added

• Initial release of Ollama Conversation integration
• Full Ollama API support with tool/function calling
• UI-based configuration flow
• Support for light control (on/off, brightness)
• Support for climate control (temperature setting)
• Conversation history management (last 10 messages)
• Configurable parameters (temperature, context_window, top_p, top_k)
• Comprehensive error handling
• Model selection dropdown
• Connection validation
• Unit tests and functional tests
• Complete documentation

Features

• Local AI processing via Ollama
• Natural language device control
• Multi-turn conversation support
• Tool execution with result feedback
• Home Assistant service integration
• HACS compatibility

Documentation

• Detailed README with usage examples
• Implementation summary
• Troubleshooting guide
• Architecture documentation
• API endpoint documentation