v1.5.0

This release brings major UX improvements to the chat interface, smarter project session handling, and critical stability fixes for custom local models and Docker networking. A standalone Windows executable is now also available.

✨ New Features & UI Enhancements
Inline Tool Approvals: Removed blocking full-screen modals. Tool execution approvals (e.g., file writes, terminal commands) are now seamlessly integrated directly into the chat feed as non-intrusive inline actions.
Interactive File Cards: File operations (read_file, write_file, etc.) now generate clickable HTML cards in the chat history. Clicking a card instantly restores the file context into the Editor overlay.
Persistent Project Sessions: Switching workspaces from the UI now automatically resumes the most recent chat session specific to that project. The Editor panel is also cleanly reset upon switching workspaces.
Markdown Reports Enforcement: The core AI agent is now instructed to strictly format detailed analyses and reports within ```md blocks, ensuring they are automatically extracted and rendered correctly in the Editor panel.
Standalone Windows Executable: Added support for compiling the entire backend and frontend into a single CoderAI_Workspace.exe using PyInstaller.
🐛 Bug Fixes & Stability
Fixed Infinite Tool Loops (LangChain Patch): Resolved the critical "Agent stopped after 20 tool iterations" bug. Patched agent_runtime.py to guarantee strict tool_call_id mapping between AIMessage and ToolMessage, preventing the agent from hallucinating tool retries when using streaming custom models or third-party API proxies.
Docker Custom API Routing: Fixed connection timeouts when using custom local APIs inside Docker. The application now intelligently auto-routes 127.0.0.1 and localhost URLs to host.docker.internal when running in a containerized environment.

RAG & memory improvement

🚀 What's New

1. 🌐 Interactive Architecture & Functions Graph (Jump to Code)

• Comprehensive Visual Topology: Renders multi-layer relationships across directories (contains), inter-file imports (imports), and execution calls (calls).
• Direct Code Navigation: Click or double-click any module, function, or class node to instantly open the corresponding file in the code editor, auto-scrolling directly to the exact definition line with syntax highlighting.
• Fluid Canvas: Fullscreen toggle on background double-click, smooth zoom and drag navigation, and rich tooltips.

2. 🧠 Local embeddinggemma RAG Integration

• Automatic Model Discovery: Scans local Ollama instances on startup for the embeddinggemma model.
• One-Click Background Pull: Interactive dark-themed popup to trigger ollama pull embeddinggemma in the background with a real-time progress bar.
• High-Precision Semantic Retrieval: Computes 768-dimensional dense vector embeddings stored locally in SQLite and Chroma for instant, accurate codebase search.
• Graceful Fallback: Automatically preserves keyword/BM25 indexing if the user chooses simple mode.

3. 💻 Real Integrated xterm.js Terminal (VS Code Style)

• Native Windows PTY (winpty): Full interactive terminal experience powered by xterm.js.
• PowerShell & CMD Support: Switch dynamically between PowerShell and Command Prompt.
• VS Code Polish: Vertical resize handle, hotkey toggle (Ctrl + \``), command history (Up/Down arrows), clear screen (Ctrl + L), and process interruption (Ctrl + C` / Kill button).

4. 🌍 Bi-directional RTL / LTR Persian & English Support

• Automatic Language Detection: Chat dynamically detects Persian and Arabic text and aligns it Right-to-Left (RTL) with proper typography, while keeping English and code blocks cleanly Left-to-Right (LTR).
• Interactive Slash (/) Commands: Rich autocomplete popover to browse and inject engineering, architecture, and productivity skills directly into conversations.

5. 📦 Single-File Windows Portable Executable

• Built with PyInstaller into a standalone executable (CoderAI_Workspace.exe).
• Zero Python setup required—just download, run, and start coding.

---

📥 Download

Asset	Description
CoderAI_Workspace.exe	Standalone portable executable for Windows (x64)
Source Code (zip / tar.gz)	Full source code with Python dependencies

---

🛠️ Requirements

• Windows 10/11 (64-bit)
• Ollama running locally (recommended models: qwen2.5-coder, gemma4, embeddinggemma)

RAG & memory improvement

Highlights

• Improved real-time response and code streaming
• Complete generated code synchronization with the built-in Editor
• Fixed incomplete downloadable code artifacts
• Added persistent project and session memory
• Added the ability to delete the current session and start fresh
• Improved codebase indexing and graph-aware RAG
• Better understanding of relationships between project files
• Enhanced automatic skill selection and skill usage tracking
• Added Git integration with approval-based changes and checkpoints
• Improved Custom API support with configurable model names
• Updated project dashboard and modular UI architecture
• Improved context-window and token usage management
• Added safeguards against indexing virtual environments and dependency folders
• Multiple stability, security, and runtime improvements
  Windows Portable Version
  The portable Windows executable is included in this release. It can be used without installing Python or configuring a development environment.
  Ollama must still be installed and running when using Local Ollama mode. Alternatively, a compatible Custom API endpoint can be configured from the application settings.

v1.1.0

What’s New
Graph-aware Codebase RAG
Project-level architecture summaries
Dependency graph based on imports and function calls
Hierarchical file, folder, and project summaries
Multi-hop retrieval across related files
Automatic routing between project-level and code-level questions
New get_project_overview and get_related_files agent tools
Improved skill selection, tracking, and usage reporting
Automatic exclusion of virtual environments and site-packages
Support for modern and legacy Ollama embedding endpoints
Expanded Index UI with entry points and dependency relationships
Incremental index updates after file changes

V1.0.1

This release introduces a redesigned project management experience and persistent memory system for local AI-assisted development.
Highlights
New Projects dashboard for managing local workspaces
Dedicated Project Home with Overview, Sessions, Files, Memory, Skills, and Settings
Persistent project and session archive using a central SQLite database
Resume previous coding sessions with their conversation history
Browse archived project files and open them directly in the editor
Project cards with Git status, file statistics, session count, recent activity, and agent status
Project-scoped semantic facts and user preferences
Improved context usage visibility
Updated Git integration and agent workflow
Refined developer-focused UI
Download
Download OllamaAgentWorkspace-v1.0.1-windows-portable.exe below.
The portable version does not require Python or manual dependency installation. Ollama must be installed and running when using local models. Custom OpenAI-compatible APIs can also be configured from the application.
Notes
This is an early release and may still contain incomplete features or edge cases. Feedback, bug reports, and contributions are welcome.

Initial Release

coderAI v0.1.0 - Initial Release

This is the first public version of coderAI, a local browser-based AI coding workspace.

Highlights

• Local Ollama model support with model selection.
• Custom API mode for OpenAI-compatible endpoints.
• Project workspace browser with file preview.
• Native folder selection for local projects.
• Agent chat with streaming responses.
• Tool calling for reading, writing, searching, and scanning project files.
• System prompt manager with editable prompts.
• Multi-skill selection support.
• Optional Tavily web search configuration.
• Context and memory controls for longer coding sessions.
• Download generated or selected code from the UI.

Notes

This is an early version and may still have rough edges. Future versions will improve packaging, memory handling, skill diagnostics, and safer execution controls.