ginlix-ai/LangAlpha

What it solves

LangAlpha is a "vibe investing" agent harness designed to move beyond one-shot AI financial analysis. It treats investing as an iterative, Bayesian process where research compounds over time, allowing users to maintain persistent workspaces for specific research goals (e.g., sector rotations or portfolio rebalancing) rather than starting from scratch with every prompt.

How it works

The system uses a dual-mode architecture to balance speed and depth:

• PTC (Programmatic Tool Calling) Mode: For deep research, the agent writes and executes Python code within a Daytona cloud sandbox. This allows it to process large financial datasets locally and return only the final result, avoiding token waste and context window limits.
• Flash Mode: Used for quick conversational responses, market lookups, and workspace orchestration.

It integrates a multi-tier data ecosystem (ginlix-data, FMP, and Yahoo Finance) and utilizes a persistent workspace architecture where an agent.md file tracks goals and findings across different conversation threads. It also supports an "Agent Swarm" of parallel subagents for market screening and a "Secretary" agent for background task management.

Who it’s for

It is built for investors and financial analysts who need a persistent, AI-augmented research environment to track theses, build financial models, and monitor markets iteratively.

Highlights

• Programmatic Tool Calling: Executes Python in sandboxes to handle bulk data and complex analysis without flooding the LLM context.
• Persistent Workspaces: Dedicated sandboxes with structured directories and a long-term memory store to preserve research across sessions.
• Financial Research Skills: 23 pre-built workflows for DCF models, earnings analysis, and initiating coverage reports.
• Multimodal Market Analysis: Ability to read PDFs and images, and the capacity to draw technical annotations (trendlines, Fibonacci retracements) directly on live TradingView charts.
• Automated Triggers: Supports both cron-based scheduling and price-triggered automations that fire when a stock hits a specific value.