LangAlpha 2.0 is live: an open-source financial agent, from investment research to agentic trading
Hi everyone! We’ve launched LangAlpha 2.0 and wanted to share what we’ve been building.
LangAlpha is an AI agent platform for financial markets. Describe an investment idea in plain language, and agents work together to research sources, analyze data, produce reports, and develop trade plans—with trade execution available through connected brokerage accounts.
For example: “Research semiconductor stocks showing confirmed upward trends. Compare their financials, valuations, and price action, produce a report, and keep tracking important developments.”
With Claude Code and Codex already available, why build LangAlpha?
Our focus is a finance-specific agent harness: the data, tools, memory, orchestration, and execution controls surrounding the model. Even with the same model and skill, results depend on the available data, how calculations are performed and verified, and whether research can accumulate over time.
LangAlpha’s technical design centers on six areas:
- Programmatic Tool Calling: process financial data through code. Agents call MCP tools through Python, using a sandbox to clean data, calculate indicators, and run analyses. Large datasets stay in files and the execution environment, with relevant results returned to the model. This reduces context usage and makes calculations easier to inspect.
- Parallel agents with human steering. LangGraph orchestrates tasks across subagents with independent context windows, covering research, data preparation, analysis, and reporting. Users can add constraints or redirect the work while it runs.
- Persistent workspaces for ongoing research. Data, charts, reports, and research notes remain available across sessions. Workspace memory and Skills preserve context and methods. Checkpoints, background tasks, and Redis/SSE event replay support long-running work and progress recovery after a connection drops.
- Traceable research and usable deliverables. Inspect the sources behind the work and generate PDFs, Excel workbooks, slide decks, and interactive dashboards. Chart analysis, scheduled tasks, and price triggers extend a research session into ongoing monitoring.
- Governed trade execution. Connect supported brokerages, including Robinhood, to turn research into trading actions. Connection permissions, order approvals, and execution checks are enforced by the host system. Brokerage credentials remain separate from the agent’s code execution environment.
- Extensible and self-hostable. Use different models, bring your own API keys, and extend the platform through Skills and MCP plugins. The stack combines React, FastAPI, LangGraph, PostgreSQL, Redis, and sandboxed execution, with support for self-hosting and custom data integrations.
Who is it for?
Individual investors with ongoing research needs, investment researchers and small teams, and developers building financial agent workflows.
Research coverage spans equities in the US, Hong Kong, mainland China, Japan, South Korea, Germany, India, and other markets, alongside crypto, gold, and other commodities. Data coverage and freshness depend on the connected providers; trading availability depends on the brokerage and account permissions.
Website: langalpha.ai
Open source: github.com/ginlix-ai/LangAlpha
We’d love feedback, issues, and code contributions—especially discussions around financial data reliability, context management, multi-agent collaboration, and permissions for agent-driven trading.
Top comments (0)