OctoBot

OctoBot runs on the operator’s own machine — from the Docker one-liner above, from prebuilt executables for Windows, macOS, Linux and Raspberry Pi, or as a Python package — and serves a web interface. Its tentacles repository supplies twelve trading modes (arbitrage, blank, daily, dca, dip_analyser, grid, index, market_making, remote_trading_signals, signal, staggered_orders and trading_view_signals) that decide and place orders on CCXT-supported exchanges, alongside a backtesting engine and a paper-trading simulator. The project has been developed since 2018.

The LLM parts are in the open-source tree rather than reserved for the paid cloud: Services/Services_bases/gpt_service/gpt.py calls the OpenAI SDK, defaults to gpt-3.5-turbo and exposes a custom base URL for Ollama-compatible models, and the tentacles tree carries a GPTEvaluator and a gpt_trading profile. Since February 2026 the main repository also ships packages/agents/octobot_agents, whose team/ subpackage defines manager, critic and judge roles communicating over channels with pydantic-typed LLM outputs. One boundary is worth knowing: running the ChatGPT evaluator live needs only the operator’s own key, but backtesting it relies on predictions precomputed by OctoBot’s hosted cloud service.

Three things to weigh before adopting it. Licensing is mixed — this repository is GPL-3.0, while the OctoBot-Tentacles, -Trading, -Commons and -Services repositories are LGPL-3.0. Release channels have diverged: GitHub releases are on a 3.0.0 beta line that defaults to a “node mode” acting as backend for a hosted UI, with the classic self-hosted interface reached by starting a manual OctoBot, and the README warns the beta is a work in progress with possible bugs and incomplete features — meanwhile PyPI still carries 2.1.1, so pip install OctoBot and the release binaries are not the same version. And the project is commercially funded: its README carries exchange affiliate referral links and markets a paid cloud service and a commercial license.