What is LangAlpha?
LangAlpha applies the coding-agent pattern to investing. Each thesis gets a workspace with files, notes and memory; a lead agent splits questions across subagents that read filings and market data in Python, and built-in skills produce DCF models, comps, initiation reports, decks and dashboards.
It connects to Robinhood, Interactive Brokers, moomoo and Webull, with live orders on the first three. Orders go through one governed path: you choose per connection what is allowed, and an approval is bound to a single call with unchanged arguments.
Ginlix AI started the Apache-2.0 repository in January 2026. It runs as a hosted beta at langalpha.ai or self-hosted with Docker, works with Claude, GPT, open-weight and local models, and desktop v0.3.2 shipped on 14 November 2026.
Is LangAlpha worth it?
The research harness is the strong part. Workspaces persist across sessions, an agent.md file tracks goals and findings, and data tools return tables to Python rather than pasting rows into the context, which matters when ten years of daily bars for a dozen tickers runs to about 30,000 rows. A Sources panel records every filing, data call and web page the agent touched, so you can check the work. Thirty-five skills ship built in, several adapted from Anthropic's financial-services plugins.
Order handling is better thought out than in most trading agents. Order tools are never exposed to sandbox code, live orders wait for approval by default, and an approval token is tied to one call and a hash of its arguments, so a changed or replayed order fails at the relay. Broker credentials are attached on the host and never enter the sandbox.
Self-hosting undercuts that. The README says the self-hosted build has one local user with no sign-in, so anyone who can reach it acts as you, connected brokerages included, and the web app and API listen on every network interface. Quotes come from Yahoo Finance or FMP every 60 seconds and price triggers poll every 30 seconds, so automations run on delayed data. The hosted beta fixes the data and the sign-in, at $20 to $200 a month.
The project is ten months old with 2,315 stars, and the pricing page carries the usual line that nothing is investment advice. TradingAgents and Vibe-Trading are simpler to read; FinceptTerminal covers more data. None of them governs orders as carefully.
Where it earns its place
An independent equity analyst
Keep a thesis on one stock current through earnings and filings
Create a workspace per company, let subagents pull filings and build a reverse DCF, and schedule a weekly assumption check. Hosted Plus at $20 a month includes 18,000 credits; your own model keys use none.
A two-person family office
Produce initiation-style reports without a junior analyst
Use the initiating-coverage and comps skills to turn workspace research into a PDF and HTML report with charts, then check every figure in the Sources panel before it leaves the office.
A developer with an Interactive Brokers account
Let an agent stage trades without letting it trade freely
Connect IBKR with staged orders only, so each order waits for confirmation in the broker's app. Run it hosted or on a machine reachable only through a VPN, since self-hosted has no login.
Before you commit
- Setup
- git clone, make config (a wizard for model, data, sandbox and search), make up; the web app is on localhost:5173 and the API on port 8000. Hosted: sign up at langalpha.ai.
- Integrations
- Robinhood, Interactive Brokers, moomoo and Webull; MCP servers as Python modules; Daytona, E2B, Docker or SSH sandboxes; Tavily, Serper, Exa and others for search; Slack, Discord, Telegram, Feishu and iMessage on the hosted service.
- Data and privacy
- Self-hosted data stays in your Postgres and object storage; keys and OAuth tokens are encrypted with pgcrypto and secrets are redacted from tool results before they reach the model. Prompts go to the model provider you choose.
- Cost at scale
- Hosted Pro is $40/month for 38,000 credits; Max is $100 or $200/month. Self-hosted costs are your model provider plus any FMP, sandbox or search keys.
- Limits
- Self-hosted runs only while your machine is up, with one unauthenticated user; options chains are hosted-only; hosted Free allows 5 workspaces and 1 automation.
What’s good
- Persistent workspaces and memory per thesis
- Data handled in Python, not pasted into context
- Order approvals bound to one call and its arguments
- Works with Claude and ChatGPT sign-in, API keys or local models
What’s not
- Self-hosted build has no sign-in and listens on all interfaces
- Self-hosted quotes are delayed and triggers poll every 30 seconds
- Options data and messaging channels are hosted-only
- Young project; hosted plans and features still in beta
Best for
Analysts and technically minded investors who want an agent to keep long-running equity research in one place, with any live trading kept behind approvals.
LangAlpha alternatives
LangAlpha: common questions
- Is LangAlpha worth it?
- LangAlpha is the most carefully engineered open agent for stock research we have read, and its self-hosted mode is too loosely secured to point at a live brokerage account. We rate it 3.6 out of 5.
- How much does LangAlpha cost?
- Self-hosted free (Apache-2.0); hosted Free 1,000 credits/month, Plus $20/month, Pro $40/month, Max $100 or $200/month; 15% off yearly; your own model keys use no credits.
- What are the downsides of LangAlpha?
- Self-hosted build has no sign-in and listens on all interfaces; self-hosted quotes are delayed and triggers poll every 30 seconds; options data and messaging channels are hosted-only; young project; hosted plans and features still in beta.
- Who is LangAlpha best for?
- Analysts and technically minded investors who want an agent to keep long-running equity research in one place, with any live trading kept behind approvals.
- How much does LangAlpha cost for a team or in production?
- Hosted Pro is $40/month for 38,000 credits; Max is $100 or $200/month. Self-hosted costs are your model provider plus any FMP, sandbox or search keys.
- Is LangAlpha safe for company data?
- Self-hosted data stays in your Postgres and object storage; keys and OAuth tokens are encrypted with pgcrypto and secrets are redacted from tool results before they reach the model. Prompts go to the model provider you choose.
- What are the best LangAlpha alternatives?
- The closest alternatives we have reviewed are TradingAgents, Vibe-Trading and Fincept Terminal.
Answers checked Nov 16, 2026.