Deep Research Agent — goes beyond simple Q&A to conduct multi-step analysis: gathering data, running technical indicators across timeframes, cross-referencing fundamentals, and producing a structured investment thesis with confidence scoring.

Chart Pattern Detection — 20+ candlestick patterns and 15+ chart patterns detected automatically with visual annotations and buy/sell signals.

Backtesting Engine — test any strategy against historical data with performance metrics, drawdown analysis, and risk-adjusted returns.

Trade Strategy Analysis — AI evaluates strategies across multiple market conditions, comparing performance against benchmarks.

Private LLM support — route all AI calls to your own GPU (Ollama, LM Studio, llama.cpp, vLLM) for complete data privacy. Your trading data never leaves your hardware.

Skills as first-class citizens — each skill is a YAML file describing inputs, outputs, validation rules, and execution logic. Skills are versioned, testable, and reusable across agents.

Dynamic workflow composition — the agent reads available skills at runtime and composes multi-step workflows on-the-fly based on the user's query. No static workflow graph needed.

Consistent, auditable results — because execution follows validated skill templates, outputs are predictable and every step is logged. You can trace exactly which skills were used and why.

Significant reduction in hallucination — in internal testing, skills-driven execution dramatically reduces errors because the agent selects from known-good patterns rather than generating novel code.

Token efficiency — skills are compact YAML references, not full code blocks. The agent sends skill names + parameters, not entire implementations.

4 task execution modes in one system:

🗣️ Prompt tasks — AI answers questions on a schedule ("Summarize market news every morning")
💻 Code generation tasks — AI writes and runs Python code dynamically ("Calculate portfolio beta weekly")
🐙 GitHub job tasks — Trigger and monitor CI/CD jobs or scripts from a repo
🔄 Workflow tasks — Chain multiple steps with conditional logic

Unlike OpenAI's operator or similar tools, this runs on your own infrastructure with full control over data, models, and execution environment. The Slack bot integration means the AI joins your team channel as a member — it can be @mentioned, respond to questions, and proactively post scheduled results.

7 specialized subagents in one system — workspace manager, project manager, coding specialist, documentation writer, testing expert, debugging assistant, and general-purpose agent. Each has its own context, skills, and delegation criteria. The triage agent routes tasks intelligently.

Project context persistence — QWork maintains a living understanding of your codebase, architecture decisions, and team conventions across sessions. Ask "why was this designed this way?" and it knows.

Template variable system — 15 context variables (project name, workspace path, session state, etc.) flow automatically through all agent prompts, eliminating repetitive context-setting.

MCP (Model Context Protocol) integration — pluggable tool servers including a stock market MCP service, enabling agents to call external APIs as naturally as reading files.

90%+ code reuse across interfaces — CLI, web UI, and VS Code companion all share the same core package. One codebase, three interfaces, consistent behavior.

Private LLM first — built around Qwen models (local) with OpenAI-compatible API support. Your code never leaves your machine.