Vibe‑Trading open‑source AI agent released

A new open-source project called Vibe-Trading packages an artificial-intelligence trading agent into a GitHub repo and Python package, with an MIT license and command-line install. (github.com) (pypi.org) The repository says version 0.1.0 landed on April 1, 2026, and the maintainers pushed follow-on updates through April 10, including multi-market backtests, TradingView export, more data connectors and 236 unit tests. (github.com 1) (github.com 2) The basic idea is simple: instead of hand-coding every rule, a user writes a request in plain English and the software routes it to built-in finance tools, data feeds and specialist agents. The project describes that workflow as turning natural-language prompts into strategies, research and portfolio analysis. (github.com) (pypi.org) That pitch lands as more developers try “agent” software that can plan steps, call tools and write code on its own. In trading, the appeal is speed: one system can screen assets, backtest an idea, compare allocations and summarize the results in one session. (github.com) (pypi.org) Vibe-Trading says it ships with 64 finance skills and 29 prebuilt “swarm” teams, which are groups of specialized agents assigned to jobs like earnings research, crypto market review or risk committee sign-off. The listed skills span technical analysis, factor research, options, macro analysis, sentiment and portfolio allocation. (pypi.org) (github.com) The package also bundles portfolio construction methods such as mean-variance optimization, risk parity, Black-Litterman and hierarchical risk parity. In plain terms, those tools try to spread money across assets by balancing expected return, volatility or correlation rather than picking a single stock. (pypi.org) On data, the project says Hong Kong, United States and crypto market workflows can run without paid keys, using Yahoo Finance and OKX, while China A-share queries need a Tushare token. The maintainers also say 15 of 16 Model Context Protocol tools work with zero API keys, and multi-agent swarm mode is the main feature that still needs a language-model key. (github.com) The interfaces are broader than a single script. After installation, users get an interactive command-line and text interface, a web server command and a Model Context Protocol server for connecting the toolkit to other assistants and developer tools. (github.com 1) (github.com 2) The repo had 931 GitHub stars and 188 forks when it was fetched, a sign of early developer interest rather than evidence that the strategies work in live markets. The project materials emphasize backtesting and research, and they do not present audited live performance results in the repository pages reviewed here. (github.com 1) (github.com 2) For now, Vibe-Trading looks less like a finished hedge fund in a box than a fast-moving toolkit for people who want to test how far artificial-intelligence agents can go in market research and automation. The next question is whether users keep it in the lab for experiments or push it into live trading with real money. (github.com) (pypi.org)