mhdashikofficial/tenshi-ai-agent

Tenshi AI Agent is a Python program that trades financial assets automatically, using an AI language model to read market conditions and decide what to buy or sell. The agent runs on its own once started, looping through market data, consulting an AI for decisions, and placing orders without human input. It is designed to run continuously on a remote server so that it can trade around the clock. The key idea behind the project is that it does not just ask the AI what to do and hope for the best. Instead, the code wraps every trade with hard-coded rules written in Python. If a position drops by 2 percent or gains 10 percent, the program forces an immediate close regardless of what the AI says. This means the AI cannot accidentally ignore a stop-loss rule the way a purely conversational system might. Risk limits are enforced by code, not by the model's reasoning. To keep costs down, the agent does not send raw market data to the AI. It pre-processes the feeds in Python, extracts the most important signals, and sends only a short summary. According to the README, this cuts the number of AI tokens used by over 80 percent compared to passing full data dumps. The agent supports several AI providers, including Google Gemini and models accessed through OpenRouter or a locally running Ollama setup. Results and activity are broadcast to a Telegram channel automatically. The agent sends live trade notifications, hourly market summaries, and daily balance reports in a readable format. It also saves its conversational memory to a file so that context is preserved if the program restarts. Setting it up requires a Python 3.9 environment, API keys for a broker (the README references an AI4TRADE token), at least one AI provider key, and a Telegram bot token. The README includes a deployment script that pushes updates to a remote server. The codebase is split into folders for the core AI loop, trading and risk logic, and utility functions for data processing and Telegram formatting.