Release v2.1

[mr-engineer49]

What's Changed

• Delete .gitattributes by @mr-engineer49 in Delete .gitattributes #1
• Refactor Python publish workflow for clarity and updates by @mr-engineer49 in Refactor Python publish workflow for clarity and updates #8
• fixing the flak8 issue by @mr-engineer49 in fixing the flak8 issue #9
• Update production.yml by @mr-engineer49 in Update production.yml #10
• Development by @mr-engineer49 in Development #11
• Create SECURITY.md for security policy by @mr-engineer49 in Create SECURITY.md for security policy #13

Changelog — Implemented Features & Modifications
Core Engine
Modular architecture — refactored from single-file (600 lines) to multi-module package (nexus/core/, nexus/ui/panels/)
Panel hibernation system — activate() / deactivate() lifecycle on every panel to save resources
QThread workers — all subprocess, LLM, and network operations run in background threads (no UI freezing)
WorkflowEngine (engine.py) — DAG-based execution engine with step-by-step debug mode, pause/resume, and context variable tracking
CodebaseRAG (rag.py) — local vector store indexing with multi-provider embedding support
LangChain Agent Engine (langchain_agent.py) — multi-provider LLM factory, 14-tool registry, LangGraph ReAct agent
Theme system (style.py) — centralized dark theme tokens and global Qt stylesheet
Settings persistence (config.py) — all settings saved to ~/.nexus_settings.json
LLM Providers
Ollama (local) with auto-detection and model listing
OpenAI with configurable base URL
Anthropic Claude
Google Gemini (langchain-google-genai)
Groq
Together AI (OpenAI-compatible endpoint)
Generic OpenAI-compatible endpoints
Agent Tools
shell — shell command execution with timeout and approval
read_file / write_file / list_dir — file system operations
git — git sub-command runner
python_repl — sandboxed Python code execution
web_search — DuckDuckGo search integration
code_search — RAG-powered semantic codebase search
http_request — HTTP GET/POST to any REST API with approval
json_transform — JSON parsing with dot-path traversal
regex_extract — regex pattern extraction
summarize_text — multi-provider text summarization (uses active LLM, falls back to Ollama)
langsmith_runs / langsmith_datasets — LangSmith query tools
UI Panels
Status Dashboard — real-time system metrics with auto-refresh hibernation
Ollama Manager — model CRUD, pull, run, chat, GPU detection
Git Panel — diff viewer, commit builder, branch/stash/tag management
GitHub Panel — repo browser, clone, star, search, create
Chat Agent — streaming AI chat with multi-provider support
Agent Studio — custom ReAct agent builder with tool selection
LangSmith Panel — observability dashboard, run inspection, dataset browser
Integrations Hub — live detection of LLM providers, dev tools, AI frameworks
Project Runner — project management with git quick-actions
RAG Search — codebase indexing, semantic search, AI Q&A with source citations
Prompt Templates & Sandbox — template library, dynamic variable forms, streaming sandbox with metrics
Visual Workflow Builder — node graph editor with DAG execution engine
Workflow Debugger — step-by-step execution, pause/highlight, runtime variable editing
Terminal — multi-shell support with history and quick-launch buttons
System Monitor — CPU/RAM/disk/swap gauges with top processes
Settings — comprehensive configuration for all providers and features
Bug Fixes & Stability
Fixed RAGIndexerWorker / RAGQAWorker missing stop() methods — prevented crash when switching panels during active operations
Fixed UI freezing during panel switches — all long-running operations moved to QThread workers
Fixed summarize_text tool hardcoded to Ollama — now uses the active LLM provider with graceful fallback
Added blockSignals() guard in workflow context table to prevent infinite loops during population
Made context variable keys read-only in the debugger table to prevent accidental key corruption

Full Changelog: release-v1...release-v2

---

This discussion was created from the release Release v2.1.