An open-source (Apache 2.0) AI agent for investment research published by Ginlix AI. The GitHub repository describes it as “Claude Code for Finance,” and the design carries over the patterns proven in coding agents — a persistent workspace where work is left on disk, parallel sub-agents, and tools that are discovered only when needed — and applies them to equity and filings research. Rather than answering in a single exchange, it is built around accumulating findings in a workspace and updating your investment view each time new material arrives. It can be used in two ways: a self-hosted setup you run on your own machine or server with Docker, and a hosted version operated by Ginlix AI (langalpha.ai).
Key Features
- A persistent workspace where research compounds: Each investigation creates directories such as
work/,results/, anddata/along with an agent notes file, so context carries across sessions. You can build on a previous analysis and revise your view as new earnings or news arrive - Parallel sub-agents working simultaneously: The main agent spawns multiple sub-agents asynchronously, assigning price data, SEC filings, valuation, and other tasks in parallel. Each sub-agent keeps its own isolated context, so reasoning is less likely to blur as the scope grows
- Python processing to save tokens: Instead of feeding large time-series data straight into the model, the agent writes Python code, aggregates and transforms the data inside a sandbox, and takes back only the results (Programmatic Tool Calling). Even years of price history put less pressure on the context window
- Layered access to financial data: Prices, fundamentals, macro indicators, options, and news are retrieved via MCP. Data providers fall back in the order ginlix-data → FMP → Yahoo Finance, so it runs at a minimum through Yahoo Finance even with no API keys at all — with the corresponding limits, such as no intraday bars and delayed quotes
- Bundled research procedures (skills): Twenty-three ready-made procedures are included, covering DCF models, earnings analysis, morning notes, and document generation, so routine research can be run without assembling prompts from scratch
- Scheduling and deliverable export: Time-based automation via cron and price-triggered runs are supported (price triggers assume the hosted version’s real-time feed). Output can be written to Word, Excel, PowerPoint, and PDF, and the agent can annotate interactive charts
Pricing
| Delivery model | Price | What it covers |
|---|---|---|
| Self-hosted (open source) | Free (Apache 2.0) | Built with Docker. You supply your own LLM and data API keys |
| Hosted (langalpha.ai) | Start free | Cloud environment and data infrastructure included. Bring your own LLM key (BYOK) |
| Hosted paid plans | Check with the provider | Higher monthly limits and top-up credits |
The official site states that the hosted platform can be started for free and that paid plans add higher monthly limits and top-up credits, but as of August 2026 no specific prices are published. Note also that with either the self-hosted or hosted option, LLM usage fees (Claude, GPT, Gemini, and so on) are charged separately.
Pricing information is current as of August 2026. Please check the official site for the latest details.
Pros & Cons
✅ Pros
- Being open source, you can read the code to confirm its behavior and run it entirely within your own environment — an easy choice if you would rather not hand your research targets or holdings to an outside service
- Results remain as files, so you can trace back why a conclusion was reached. That is a different character from disposable chat
- Parallel sub-agents and Python preprocessing keep things from breaking down as the volume of data grows
- You can try it with no API keys at all through Yahoo Finance, which lowers the barrier to getting started
- Reports can be exported to Word, Excel, PowerPoint, and other formats used in day-to-day work
⚠️ Cons
- Setup requires knowledge of Docker. This is not a tool a complete beginner can use right away
- If you want high-quality data, a paid data contract such as FMP is effectively necessary; a free configuration leaves gaps in intraday bars and macro indicators
- LLM usage is paid by the user, so running large-scale research accumulates model-side cost
- The documentation explicitly states that the Docker sandbox offers weaker isolation than the cloud sandbox (Daytona)
- As the project states plainly that it is a research tool and not a financial advisor, verifying the output is the user’s responsibility
- Specific prices for the hosted paid plans are not published, which makes cost estimation difficult
Comparison with Similar Services
| Item | LangAlpha | OpenBB | General AI chat (Claude / ChatGPT, etc.) | DIY Claude Code + finance MCP |
|---|---|---|---|---|
| Positioning | Finance-specific AI agent platform | Data platform for investment research | General-purpose conversational AI | Repurposed general coding agent |
| License | Open source (Apache 2.0) | Primarily open source | Proprietary | Commercial harness, self-assembled setup |
| Research continuity | Accumulated in a workspace | Retained as an analysis environment | Per conversation or project | Accumulated in a workspace |
| Financial data access | Layered fallback out of the box | Centered on broad data integration | Depends on search or connectors | You assemble the MCP yourself |
| Ready-made procedures | 23 bundled skills | Toolkit-centered | You write your own prompts | You write your own |
Who Is It For
- Individual investors who research specific stocks continuously and want to stop starting from zero every time
- Analysts who must follow several names in parallel during earnings season and want to automate the groundwork
- Engineering-minded users who want everything to stay in their own environment, or who want to verify behavior at the code level
- Developers curious about examples of applying an agent like Claude Code to domains outside coding
- Anyone who needs to share research reports with colleagues in Word or Excel form
Summary
LangAlpha is an open-source agent platform that carries the design philosophy of coding agents — work left as files, sub-agents running in parallel, heavy processing pushed into code — directly into financial research. It fits a research style of following names over time, but it comes with prerequisites: a Docker build, data contracts, and LLM usage fees, so it is not a tool to pick up casually. A realistic path is to run the self-hosted version on the Yahoo Finance configuration first to get a feel for it, then consider data providers or the hosted version when moving to serious use. As the project’s own notice says, it is a tool to assist research, not something whose output should be turned directly into investment decisions.