Author runs a self-built autonomous agent on Codex, ChatGPT Plus, systemd and email that handles various company tasks, using a rolling context, diary and markdown wiki for memory.
Author runs a self-built autonomous agent on Codex, ChatGPT Plus, systemd and email that handles various company tasks, using a rolling context, diary and markdown wiki for memory.
The author built their own autonomous agent instead of using OpenClaw. It runs on Codex with a ChatGPT Plus account, is scheduled by systemd, and communicates through ordinary UNIX email and a normal user account. For memory it uses one long rolling context window, a diary, and a folder of markdown wiki files. It works as an employee of the author's company.
The author reports it works great and has handled all kinds of tasks, with no memory problems so far. They say it is early days and demands may grow. Specifics on tasks and results are not given.
If memory strains as workload grows, the author's plan is to add a second or third bot as specialists, like employees, rather than overloading one.
Help me build an always-on agent for my company that handles [recurring tasks, such as triaging email or updating docs]. It should run as a [systemd service or scheduled job], keep memory in a markdown wiki and a daily diary file, and read tasks from [inbox or folder]. Finished means it completes one test task end to end and logs what it did. Require my approval before it sends any email or deletes anything.
Export the Descript transcript, let ChatGPT pick the five strongest moments, and get 10 platform-ready posts in a review doc within minutes.
Give it a keyword and get a brief built from the pages that rank today: headings, gaps, questions to answer and a target length.
Spot order-status questions in the support inbox, look up the order in Shopify and reply with live tracking and an honest ETA.