ValueCell

Three agents ship: a DeepResearch agent over fundamental documents, a Strategy agent that supports multi-asset, multi-strategy crypto trading and automatically executes the strategies it produces, and a news retrieval agent for scheduled delivery. Underneath is an async, re-entrant Python orchestrator with human-in-the-loop pauses and first-class Agent2Agent support through a2a-sdk, with a documented path for contributing new agents. It runs self-hosted as a Python and React app with a web UI, and signed macOS and Windows desktop builds are published — all release artifacts to date are labelled beta.

Live trading is perpetual contracts only: the README states that spot is implemented as 1x contracts, so the connected account must be a funded perps account. Binance (international site, USDT-M futures), OKX and Hyperliquid are marked tested, while Coinbase, Gate.io, MEXC and Blockchain are only partially tested. Hyperliquid carries its own constraints — USDC margin only, market orders auto-converted to IoC limit orders, and a 10 USDC minimum per trade.

Two things to weigh. Apache-2.0 does not cover everything shipped: the README states it applies only to original code authored by the ValueCell team and contributors, while the embedded TradingView Advanced Chart widget is proprietary under its own agreement and exchange and LLM APIs carry their own terms. And the project is dormant at the maintainer level — no commit on main and no merged pull request since February 2026, and no release since v0.1.20 in January 2026, while community pull requests continue to arrive unmerged. A hosted product at valuecell.ai is promoted at the top of the README; the self-hosted path is complete and no pricing, token or returns claim appears there.