A Self-Hosted AI Research Assistant That Keeps Your Portfolio Data on Your Own Machine

You've probably tried pointing a general-purpose chatbot at a stock question and gotten something that sounds confident but has no idea what a P/E ratio is. Or maybe you've looked at hosted financial AI tools and balked at shipping your positions, watchlists, and investment theses off to someone else's server. HunterCode is a self-hosted answer to both problems: an AI investment research assistant that runs on your own hardware, keeps its data local, and works across A-shares, Hong Kong stocks, and US equities.

What It Does

HunterCode is a financial AI assistant you run yourself, either on your laptop or a server you control. You give it a large language model API key, and it handles market data lookups, news retrieval, per-stock deep analysis, trend prediction, and management of your watchlist and holdings. It follows a methodology you define yourself through something the project calls a SKILL. Conversations, positions, and investment theses all live in a local database rather than on someone else's infrastructure.

Under the hood it's a multi-service application (the README references six services, Postgres, and Redis) that ships as Docker-based deployments. There's a web UI, an onboarding wizard, and a desktop launcher for people who'd rather not touch a terminal. The project positions itself as an open-source, local alternative to Tencent's WorkBuddy finance edition, aimed at private funds and serious individual investors. It's Apache 2.0 licensed.

To be clear about what it isn't: it doesn't connect to a broker for trading, it doesn't make decisions for you, and it doesn't promise accurate predictions. It's a tool that organizes public data, your analysis methodology, and an LLM into a research workflow.

Why It's Cool

The data locality is the whole point. Every conversation, every position you log, every thesis you write stays in a database on your machine. For anyone managing real money, that's not a nice-to-have, it's the reason to use this over a hosted product. You're not trusting a vendor with your book.

The SKILL concept is the smart part. Rather than hardcoding one analysis framework, HunterCode lets you write your own methodology and have the AI work within it. That means the tool adapts to how you actually think about markets instead of forcing you into someone else's template.

The deployment options are unusually honest. The README lists Zeabur, Sealos, Railway, 1Panel, and Coolify/Dokploy, and then does something you rarely see: it admits the maintainers haven't actually deployed to any of these platforms. There's no one-click button, just documentation. Each template was "equivalently verified" by mechanically translating it into a compose file and running the full flow locally (six services healthy, onboarding complete, real conversation, deep analysis, restart with data intact). Each doc spells out what was and wasn't tested. If you have an account on one of these platforms and test it, they'll add the button.

Security got real attention. Public deployments default to disabling single-user no-login mode and generate a HUNTER_SETUP_TOKEN you enter during onboarding. Without it, whoever opens the instance first could configure the LLM with their own key. Small detail, but it shows they thought about what happens when someone exposes this to the internet.

There's a desktop launcher for the command-line-averse. Download, paste a key, wait a few minutes. It checks for Docker, picks a download source, finds a free port, writes config, pulls images, and starts containers, then opens your browser. That's a meaningful accessibility win for analysts who aren't developers.

How to Try It

The fastest path is the desktop launcher, available for macOS 12+ (universal Intel/Apple silicon) and Windows 10/11 x64. Grab it from the project's download page or GitHub Releases, run it, and follow the wizard: welcome, paste key, authorize, pick a model, auto-install.

If you'd rather deploy to a cloud platform, the repo has per-platform guides:

• Zeabur (works domestically and abroad, most capable)
• Sealos (for users in China, K8s template)
• Railway (overseas, manual six-service setup)
• 1Panel (own server with a domestic panel)
• Coolify / Dokploy (own server, two paste-ready compose files)

For a manual local install, start with the quick start docs:

https://github.com/agentpit-io/hunter-community

There's also an online demo at hunter-community.agentpit.io if you want to poke around before installing anything, and the full product manual is in the repo's doc directory.

Final Thoughts

HunterCode is best suited to people who already use Docker, care about where their financial data lives, and want to plug their own data sources and methodology into an AI workflow. The honest caveats in the README (no broker integration, no accuracy guarantees, deployment templates unverified on real platforms) are refreshing, but they also mean this isn't a turnkey product yet. It's a serious tool for a specific audience. If you're a private fund researcher or a technically comfortable individual investor who's been waiting for a local-first alternative in this space, it's worth the afternoon.

Follow @githubprojects for more developer tools and open source projects.