An AI portfolio manager that reads your brokerage screenshots

What it does DojoAgents is an open-source Python framework that wraps a large language model in an autonomous “Agent Loop” to act as a personal investment analyst. It ingests multi-market data—A-shares, US equities, Hong Kong stocks—and connects it to your actual portfolio holdings to run structured analysis across fundamentals, momentum, sector taxonomy, and risk exposure. Results surface in a React dashboard or through chat gateways, with the agent running background research, news impact decomposition, and simulated portfolio generation.

The interesting bit The standout trick is multimodal portfolio ingestion: upload a screenshot of your brokerage app and the agent recognizes each position, re-groups them by sector and supply-chain exposure, then flags hidden concentration risks—like forty semiconductor-adjacent names masquerading as diversification. It is a genuinely clever use of vision models to skip manual data entry, though the README is vague on parsing accuracy for obscure tickers.

Key highlights

• Portfolio-aware reasoning loop that cross-references your holdings with live market data rather than generating generic commentary.
• Execution sandbox for on-the-fly Python code execution and technical indicator calculation.
• Memory and “SKILLS” system that distills successful multi-step analysis workflows into reusable templates.
• Supports both commercial LLM APIs (OpenAI, Anthropic, Gemini, DeepSeek) and local endpoints via Ollama or vLLM.
• Apache 2.0 license; explicitly framed as educational and research-only, not a trading recommendation engine.

Caveats

• Omnichannel alerts (Slack, Telegram, Discord, WeChat, email) are listed as “planned” for future versions, not shipped yet.
• The project disclaims itself as educational and non-advisory; all outputs are explicitly “for reference only.”
• The README leans heavily on aspirational language (“state-of-the-art,” “institutional-grade”) without publishing accuracy metrics or benchmark comparisons.

Verdict Worth a look if you are a retail developer-investor who wants to experiment with autonomous agents over your own portfolio data. Skip it if you need audited, production-grade trade execution or a fully offline appliance—this is a research framework with a dashboard, not a certified robo-advisor.