Alpha-Dojo/DojoAgents

DojoAgents – AI‑powered personal‑investment copilot

DojoAgents is an open‑source framework that lets individual investors run a portfolio‑aware AI agent on their own machine. The core is an Agent Loop that orchestrates a large‑language model (OpenAI, Gemini, Anthropic, etc.) together with a set of finance‑specific tools (market data fetchers, news scrapers, technical‑indicator calculators, Python sandbox, etc.). The agent can reason over multiple markets (US, Hong‑Kong, A‑shares), keep track of the user’s holdings, and produce multi‑step analyses such as daily market overviews, news‑impact assessments, or risk diagnostics from a screenshot of a portfolio.

Key components

Component	What it does
Agent Loop Engine	Runs the LLM, decides which tool to call, compresses context, and enforces guardrails to avoid “financial hallucinations”.
Execution Sandbox	Isolated Python environment for on‑the‑fly calculations (e.g., technical indicators) and web scraping.
Memory & SKILLS	Stores successful multi‑step workflows as reusable “skills” that the agent can invoke later.
Cron & Gateway	Schedules automated insights and pushes them to chat platforms (Slack, Telegram, Discord, WeChat, etc.).
FastAPI backend + React SPA dashboard	Provides a web UI where users can view portfolio metrics, market heatmaps, sector trees, and interact with the agent via chat.

What you can do with it

Daily market overview – ask “What’s worth watching today?” and the agent gathers cross‑market movers, sector rotations, and price/volume shifts, showing each tool call it made.
News‑impact analysis – feed a headline (e.g., Meta’s AI‑infrastructure move) and receive a breakdown of which US and A‑share stocks could be affected, with supporting data.
Portfolio diagnosis from a screenshot – upload a picture of a holdings list; a multimodal model extracts the tickers, groups them, and flags concentration or missing defensive assets.
Simulated rebalancing – after diagnosis, the agent suggests trades, creates a “what‑if” portfolio, and tracks its performance in the dashboard.

Quick start (from the README)

# 1. Create a virtual env with uv (recommended) and install the package
uv venv && source .venv/bin/activate
uv pip install dojoagents   # pulls the published wheel

# 2. (Optional) Build the web UI from source
cd dojoagents/dashboard/web
npm install && npm run build

# 3. Run the server
dojoagents dashboard --host 127.0.0.1 --port 8765

Open the shown URL in a browser, then configure your LLM API key in the Settings panel (supports OpenAI, Anthropic, Gemini, Zhipu GLM, DeepSeek, or any local endpoint like Ollama). The UI is a React SPA that talks to the FastAPI backend via an OpenAI‑compatible chat endpoint and Server‑Sent Events for live chart updates.

Extensibility

Custom Harnesses – the framework separates the financial harness (prompts, tools, policies) from the generic runtime. Developers can supply their own harness YAML or a Python factory to target other domains.
Plugins / Skills – add new data sources, write Python‑based indicator functions, or create reusable multi‑step analysis scripts that the agent can call automatically.
Omnichannel gateways – plug the agent into Slack, Discord, WeChat, Feishu, email, etc., for scheduled briefings.

License & Disclaimer

DojoAgents is released under the Apache 2.0 license. The authors stress that the software is for educational and research purposes only and does not constitute investment advice; users bear all risk for any trading decisions.

Bottom line: DojoAgents is a genuine, end‑to‑end AI‑agent platform that combines LLM reasoning, tool orchestration, and a polished financial dashboard to give retail investors an institutional‑style analytical assistant. It is fully open source, extensible, and can be run locally with any LLM provider you prefer.