LangAlpha is an AI-powered agent designed to interpret financial markets and support investment decisions through iterative research. It targets engineers, researchers, and power users seeking a persistent, compounding research environment, offering a significant benefit over traditional one-shot AI finance tools by enabling continuous refinement of investment theses.

How It Works

LangAlpha employs a novel Programmatic Tool Calling (PTC) architecture, where the agent writes and executes Python code within sandboxed environments to process financial data. This approach dramatically reduces token waste and enables complex, multi-step analyses that would otherwise exceed LLM context limits. Research is organized into persistent workspaces, each mapping to a dedicated sandbox with structured directories and a persistent memory file (agent.md) that compounds research across sessions. An agent swarm model allows parallel execution of specialized subagents, enhancing efficiency and enabling mid-execution steering for course correction.

Quick Start & Requirements

• Primary Install: Docker-based setup. Clone the repository, run make config for an interactive wizard to set up LLM, data sources, sandbox, and search, then make up to start services.
• Prerequisites: Docker, LLM subscription API keys (e.g., OpenAI, Anthropic, Kimi, GLM). Optional but recommended keys for full functionality include DAYTONA_API_KEY (cloud sandboxes), FMP_API_KEY (high-quality data), SERPER_API_KEY or TAVILY_API_KEY (web search), and LANGSMITH_API_KEY (observability).
• Resource Footprint: Requires Docker, PostgreSQL, and Redis. Setup time is minimal with Docker and the make utility.
• Links: GitHub Repo, API Docs (when running locally). A hosted version is available.

Highlighted Details

• Programmatic Tool Calling (PTC): Agent writes and executes Python in sandboxes for complex data processing, reducing token waste and enabling deep analysis.
• Persistent Workspaces: Research compounds across sessions via structured directories and an agent.md memory file within dedicated sandboxes.
• Agent Swarm & Live Steering: Parallel async subagents execute concurrently, with the ability to steer or update running agents mid-execution without restarts.
• Multi-Provider Model Layer: Abstracts across various LLM backends (Gemini, OpenAI, Anthropic, etc.) with automatic failover and normalized reasoning effort.
• Financial Data Ecosystem: Utilizes native tools for quick lookups and MCP servers for bulk data processing via PTC, with a three-tier provider fallback (ginlix-data, FMP, Yahoo Finance).

Maintenance & Community

For partnerships, collaborations, or inquiries, contact contact@ginlix.ai. The project appears actively developed, with ongoing work on the main branch beyond hackathon submissions.

Licensing & Compatibility

• License: Apache License 2.0.
• Compatibility: Permissive license suitable for commercial use and integration with closed-source projects.

Limitations & Caveats

Without external API keys (e.g., FMP, Daytona), the experience is reduced, relying on free but limited Yahoo Finance data and Docker sandboxes with downgraded security. Price-triggered automations require the real-time WebSocket feed from ginlix-data, currently exclusive to the hosted platform during beta. The project is a research tool and explicitly disclaims providing financial advice.