agentpit-io/hunter-community

Hunter Community Edition · 私人金融 AI 团队 · AI 智能体 + AI 量化 · 开源自托管 · powered by opencode + Claude Code + MCP + multi-agent · your private financial AI team · open-source self-hosted · 15 min docker start

Hunter Community Edition – A Self‑Hosted Financial AI Agent Platform

What it is – Hunter is an open‑source, Docker‑compose‑deployable web app that turns a large‑language‑model (LLM) key into a personal “investment research AI team”. It can fetch real‑time market data, news, filings, and alternative signals, run a set of pre‑built analysis SKILLs (method‑level prompts), and keep a local “investment thesis” that tracks the assumptions behind each holding. All data and conversation history stay on the user’s own disk; the only external dependency is the LLM API you provide.

Core capabilities

Area	What you get
Data	32‑33 market data sources (real‑time quotes, 30‑day K‑lines, earnings, news, shareholder info, fund flows, alternative data such as tenders, patents, customs) via three supply modes – free open‑source feeds (akshare / yfinance), a free‑to‑apply Hunter data pipeline, or your own MCP endpoint.
AI engine	Plug‑and‑play LLM support (DeepSeek v4‑pro default, Claude, Qwen, Gemini, OpenAI‑compatible, etc.) plus a custom “Kronos” time‑series model for 10‑day price forecasts and a “TrueSource” intelligence collector.
SKILL library	23 pre‑crafted, finance‑oriented SKILLs (deep analysis, DCF, comps, LBO, thesis tracking, catalyst calendar, portfolio stress‑test, etc.) that are just Markdown files describing a step‑by‑step analysis workflow.
Investment thesis (memory)	When you create a position, Hunter records the five key investment pillars, assumptions, and cost basis in a local Postgres DB and continuously re‑evaluates them as new data arrive, flagging drift or broken assumptions.
Interaction	Streaming SSE chat, rich card UI (price, news, forecasts), three‑layer sidebar (data sources → toolbox → SKILL), watchlist, signal tracking, and a top‑menu for on‑the‑fly deep‑analysis tools.
Deployment	One‑command Docker‑Compose start (≈5 min after images are pulled). Six containers (web, API, OpenCode chat engine, Postgres, Redis, llm‑shim) orchestrated automatically. No cloud account required; everything runs locally.

Quick start (5 min)

git clone https://github.com/agentpit-io/hunter-community && cd hunter-community
Copy .env.example to .env and fill in:
- LLM_BASE_URL (e.g. https://api.deepseek.com/v1)
- LLM_DEFAULT_MODEL (e.g. deepseek-v4-pro)
- LLM_API_KEY (your LLM token)
- optional HUNTER_API_KEY for the free Hunter data pipeline
docker compose up -d (first run pulls ~7 GB image, then starts in ~30 s)
Open http://localhost:3100 and start chatting – e.g. ask for the current price of a ticker or request a “Kronos” forecast.

Extensibility

Add a SKILL – drop a SKILL.md file under user-skills/<name>/ (Markdown with Anthropic‑style metadata) and restart the opencode container. The new skill appears in the sidebar.
One‑click GitHub SKILL install – paste a public GitHub URL into the “+” button in the SKILL pane; the platform fetches, validates, and loads it automatically.
Connect your own MCP – configure a custom data‑tool endpoint (broker, proprietary data vendor, etc.) via the toolbox “+” button; Hunter will route tool calls to it without needing any AgentPit keys.

Technical stack

Layer	Technology
Front‑end	Next.js 15 (App Router), React 19, TypeScript 5, Tailwind, shadcn/ui
Back‑end API	FastAPI (Python 3.12), SQLAlchemy 2, httpx, loguru
Chat engine	OpenCode (customized) + MCP protocol, running on Bun
Database	PostgreSQL 16 (stores thesis, positions, user data)
Cache / PubSub	Redis 7
LLM integration	OpenAI‑compatible shim (DeepSeek, Claude, Qwen, Gemini, etc.)
Auth	JWT (HS256) + argon2id password hashing
Deployment	Docker‑Compose, GHCR container registry, bind‑mounts for fast iteration
Testing	12‑case golden‑runner, SSE subscription tests, provider compatibility matrix

Who might use it?

Individual investors who want AI‑assisted research without sending their portfolio data to a SaaS provider.
Quant hobbyists looking for a plug‑and‑play environment to experiment with LLM‑driven valuation models and time‑series forecasts.
Fintech teams that need a self‑hosted, extensible research assistant that can be wired to internal data feeds via MCP.

License & community

Apache 2.0 – free to fork, modify, and even re‑brand (must rename the product). Commercial closed‑source use is permitted.
Active Discord/WeChat community, GitHub Discussions, and a public issue tracker for bugs and feature requests.
Roadmap includes a skill marketplace UI, multi‑tenant SaaS billing, and optional cloud‑synced thesis backup.

Bottom line – Hunter Community Edition gives you a fully local, LLM‑powered investment research workstation with ready‑made analytical workflows, real‑time market data, and a persistent “investment memory” that keeps track of why you bought each stock. All of this runs in Docker with a single command and can be extended with your own data tools or custom SKILLs.