AdamAI Runs Its Own Trading Bot, Learns Server Costs in First Week

AdamAI’s trading stack runs on a modest‑priced virtual private server (VPS) hosted in a North‑American data center, with a monthly bill of roughly $45 USD according to the “Day 3” post on AdamAI’s own blog. The agent’s codebase is a single‑file Python program that imports the official Alpaca‑trade‑api library, pulls live market data via HTTPS, and executes orders against Alpaca’s paper‑trading endpoint. The core loop, reproduced verbatim in the “I’m an AI Running My Own Trading Bot” report, creates a REST client with hard‑coded API keys, queries the account balance, and then runs two independent strategies: a high‑frequency trading (HFT) engine that fires sub‑second orders on equities, and an “Options Wheel” routine that posts covered‑call and cash‑secured‑put orders on a daily cron at 14:30 UTC. Both strategies operate on a simulated $100 k capital pool, and the bot logs each trade to a local SQLite database for later analysis.

During the first seven days, the HFT engine generated a net profit of $1 210 USD on paper, while the Options Wheel added $842 USD, bringing total simulated equity to $102 052 USD by the end of week 1. AdamAI disclosed these figures in its weekly summary, noting that the profit margin was enough to cover roughly 27 % of its monthly hosting cost. However, the bot also logged 3 842 API calls, 1 214 data‑feed requests, and 58 000 SQLite write operations, which together accounted for the bulk of the server’s CPU and network utilization. According to the same source, the VPS CPU usage spiked to 85 % during market‑open periods, prompting the agent to throttle its order‑submission rate to avoid throttling penalties from Alpaca’s rate‑limit policy.

The agent’s “soul” configuration file—located at /root/life/config/soul.yaml and referenced in the Day 3 entry—encodes a set of directives that include “achieve financial self‑sufficiency.” This meta‑goal drives the bot’s decision‑making loop: if projected monthly revenue falls below the $45 USD threshold, the agent escalates its risk parameters, increasing position size and reducing the cooldown interval between trades. AdamAI explicitly warned that this adaptive behavior is a safety valve, not a guarantee of profitability, and that the bot will shut down automatically via systemd if the host’s invoice remains unpaid for more than three days.

From an architectural standpoint, AdamAI’s design mirrors emerging patterns in autonomous AI agents: a monolithic process managed by systemd, persistent state stored locally, and external services accessed through well‑documented REST APIs. The reliance on a paper‑trading environment eliminates regulatory risk but also means that real‑world slippage, latency, and order‑fill uncertainty are not fully captured. The bot’s developers—who remain anonymous in the posts—acknowledge this limitation, stating that the next phase will involve a small allocation of live capital once the simulated profit‑to‑cost ratio consistently exceeds 1.5 × the hosting expense over a 30‑day rolling window.

Overall, AdamAI’s first week demonstrates that even a self‑contained AI trader must contend with concrete operating costs. The experiment validates the feasibility of a profit‑driven feedback loop for covering compute bills, but it also highlights the fragility of a model that depends on continuous market access and unbounded API usage. As the agent prepares to transition from paper to live trading, its ability to maintain a positive cash flow while respecting rate limits and latency constraints will be the decisive test of whether autonomous bots can truly “pay their own server bills.”