v0.4.2 - CLI Tool Support Fixes

Bug Fix Release

Fixed

• Fixed CLI tool support for all Ollama models (removed hardcoded whitelist)
• Fixed missing tool documentation in CLI system prompts
• Fixed AttributeError on Ctrl+C (session_id path correction)

Changed

• Reverted to optimistic tool binding for Ollama models with graceful fallback

See CHANGELOG.md for full details.

Consoul v0.4.1 - Hotfix

Hotfix Release - Critical Bug Fix

Fixed

🐛 Critical: Fixed TypeError in chat, ask, and resume commands

This hotfix resolves a critical bug introduced in v0.4.0 during SDK refactoring that caused consoul chat, consoul ask, and consoul resume commands to fail with:

TypeError: ChatSession.__init__() got an unexpected keyword argument 'tool_registry'

Changes

• Removed invalid tool_registry parameter from ChatSession.__init__() calls
• Added approval_provider parameter to ConversationService.from_config()
• Fixed approval provider flow: CLI → ChatSession → ConversationService → ToolRegistry

Testing

Verified fix with:

echo "content" | consoul ask --stdin "summarize"  # ✅ Works

All affected commands now function correctly.

Installation

pip install consoul==0.4.1 --upgrade

Consoul v0.4.0 - SDK Decoupling & Thinking Mode

Major Feature Release - SDK Decoupling & Thinking Mode Support

This release represents a significant architectural milestone with the completion of SDK decoupling (EPIC-012, EPIC-013), making Consoul's AI capabilities available as a standalone SDK. It also introduces thinking mode support for reasoning models and modern async streaming architecture.

Stats: 104 commits since v0.3.0 (22 features, 37 fixes, 0 breaking changes)

Added

SDK Service Layer (EPIC-013)

• 🏗️ ConversationService - Centralized conversation management with message history tracking, async streaming support, and tool execution orchestration
• 🤖 ModelService - Unified interface for all AI providers with dynamic local model discovery
• 🛠️ ToolService - Tool catalog and execution with async approval workflow

Thinking Mode Support (SOUL-280, SOUL-283)

• 🧠 Native support for reasoning models (phi4-reasoning:14b, DeepSeek-R1, Qwen QWQ, o1-preview)
• Automatic detection of thinking tags with real-time streaming display
• Collapsible thinking sections in message bubbles

Async Streaming Architecture (SOUL-279, SOUL-234)

• 🔄 Modern async/await streaming with StreamingOrchestrator
• 🌐 WebSocket streaming support with FastAPI proof-of-concept

Headless SDK Support (SOUL-251)

• ✨ stream_chunks() function for UI-agnostic streaming
• 📦 StreamChunk model with pure data (no UI dependencies)

Enhanced Model Picker (SOUL-276)

• 🎨 Card-based model picker UI with provider filtering and search

Fixed

• 🐛 IndexError in docs generator (SOUL-272)
• 🎨 Thinking mode UI improvements
• 🔧 Model switching and tool execution bugs
• 🌐 WebSocket streaming fixes (SOUL-277)

Changed

• 🏗️ TUI refactored to use SDK services (EPIC-012, SOUL-270) - Extracted 1,600+ lines to reusable SDK
• 📝 System prompt building moved to SDK (SOUL-253)
• 🎯 CLI session orchestration extracted (SOUL-252, SOUL-254)
• 📊 Model registry integration replacing static catalog

Full changelog: https://github.com/goatbytes/consoul/blob/main/CHANGELOG.md

v0.3.0 - Inline Command Execution

v0.3.0 - Inline Command Execution

Feature Release - This release introduces powerful inline shell command execution, allowing users to run shell commands directly from the chat interface and automatically include their output in AI conversations.

🚀 Major Features

Inline Shell Command Execution (SOUL-196)

Run shell commands directly from the chat interface with !-prefix syntax:

• Standalone mode: !ls -la executes command and displays output
• Inline mode: Here is the file: !cat README.md`` embeds output in message
• Support for complex commands with pipes, redirects, and arguments
• Automatic output truncation for large results
• Real-time execution with visual feedback
• Syntax-highlighted command output
• Structured context injection for LLM comprehension

AI Provider Improvements

• Anthropic prompt caching cost tracking (SOUL-186) - Track cache creation and read tokens separately
• ChatLlamaCpp context size extraction (SOUL-231) - Runtime extraction of actual n_ctx from loaded models
• Pattern-based intelligent defaults - Automatic context size detection for new AI models

TUI Enhancements

• Search configuration improvements (SOUL-183)
• InitializationErrorScreen (SOUL-192) - Graceful handling of failures

CLI Enhancements

• PDF support for --file and --glob flags (SOUL-211)
• --system-file flag (SOUL-212) - Read system prompts from files

🔧 Fixes

Security & Dependencies (SOUL-228)

• torch updated to 2.9.1 (CVE-2025-32434, CVE-2025-3730, CVE-2025-2953)
• urllib3 updated to 2.6.0 (CVE-2025-66471, CVE-2025-66418)

AI Provider Fixes

• OpenAI stream_options fix (SOUL-233) - Removed unconditional parameter for non-streaming calls

TUI Fixes

• Model field made optional in ProfileConfig
• Temperature field added to profile editor
• Markdown syntax highlighting improvements

📝 Changed

• Command output always expanded by default for better UX
• Context injection format improved with structured XML-like tags
• Model field made optional in ProfileConfig

📚 Documentation

• SDK integration guide for real-world usage
• Installation instructions updated
• CLI usage examples added to README
• TUI documentation completed
• Performance optimization documentation completed

✅ Testing

• 20 comprehensive unit tests for command detection
• Pattern matching validation
• Edge case handling

⚡ Performance

• Async command execution via thread pool
• Smart output truncation for large results
• Ollama context sizes cached

🔄 Upgrade Notes

• Backward compatible with 0.2.x
• New inline command execution requires no configuration
• Security updates strongly recommended
• Configuration: ~/.config/consoul/config.yaml
• History: ~/.local/share/consoul/conversations.db
• Ollama cache: ~/.consoul/ollama_context_cache.json

📦 Installation

pip install --upgrade consoul[tui]

See CHANGELOG.md for complete details.

v0.2.0 - First Public Release

First Public Release 🎉

This is the first public release of Consoul, featuring a complete TUI, comprehensive CLI commands, multi-provider AI support, and powerful tool calling capabilities.

Installation

pip install consoul

Quick Start

# Set your API key
export ANTHROPIC_API_KEY=your-key-here  # Claude
export OPENAI_API_KEY=your-key-here     # GPT-4
export GOOGLE_API_KEY=your-key-here     # Gemini

# Launch the TUI
consoul

# Or use the CLI
consoul ask "What is 2+2?"

Highlights

🎨 Beautiful TUI

• Rich, interactive terminal interface powered by Textual
• Multi-turn conversations with streaming responses
• Conversation history and search
• File attachments and image analysis
• Customizable themes (light/dark)

🤖 Multi-Provider Support

• Anthropic Claude - Claude 3.5 Sonnet, Opus, Haiku
• OpenAI - GPT-4o, GPT-4, GPT-3.5
• Google Gemini - Gemini 2.0 Flash, Pro
• Ollama - Run models locally

🛠️ AI-Powered Tools

• File editing with safety controls
• Code search and analysis
• Image analysis (vision support)
• Bash command execution
• Web search integration

📝 Complete CLI

• consoul ask - One-off questions
• consoul chat - Interactive sessions
• consoul history - Conversation management
• consoul config - Configuration management

What's New

CLI Commands

• 🎯 consoul ask - One-off questions with --stdin, --file, --glob, and --system flags
• 💬 consoul chat - Interactive chat with full context
• 📚 consoul history - Complete conversation management (list, show, export, search, resume)
• ⚙️ consoul config - Configuration management

Image Analysis

• 📸 Multimodal vision capabilities for Claude 3.5 Sonnet, GPT-4o, Gemini 2.0 Flash, and Ollama LLaVA
• Analyze screenshots, diagrams, UI mockups, and other visual content
• Security features: file size limits, extension filtering, path blocking

Token Counting Improvements

• 🚀 HuggingFace tokenizer integration for Ollama models
• 100% accurate token counting (vs 66% with character approximation)
• <5ms performance (vs 3-10+ seconds with API calls)

Performance

• Token counting: 3-10+ seconds → <5ms for Ollama models
• Real-time streaming across all providers

Security

• Image analysis security with path blocking
• Permission policies (paranoid, balanced, trusting, unrestricted)
• Audit logging for tool executions
• Command validation

Documentation

📖 Full Documentation

• Installation Guide
• Quick Start
• User Guide
• API Reference

Upgrade Notes

This is the first public release. Configuration is stored in:

• Config: ~/.config/consoul/config.yaml
• History: ~/.local/share/consoul/conversations.db

See the configuration guide for all available options.

---

Full Changelog: https://github.com/goatbytes/consoul/blob/v0.2.0/CHANGELOG.md