 | Ive had this setup going for about 2 months. 3 openclaw instances, each a persona in my side project team: a CTO agent i consult on engineering decisions, a CMO agent i use for growth strategy and content planning, a CEO agent that helps me prioritize when the other two disagree. screenshot of fleet view showing 3 instances, all Running, MODEL gpt-5.4, characters CEO/CMO/CTO The interesting part is all 3 run on my existing $20 chatgpt plus subscription. Openclaw is an open source framework that runs an LLM as a persistent worker, and it supports codex oauth which lets you log in with your chatgpt account and route model calls through your plus quota instead of a separate api bill. So my $20 covers all 3 since the agents run on schedules, actual token usage is pretty light and chatgpt plus handles it fine. Runner is ClawFleet, built for exactly this "chatgpt oauth in docker containers" shape. I just didnt want to manage api keys, which this week turned out to be lucky because anthropic cut off third-party tools from claude subscriptions. Kind of amazed nobody in my circle knew this was a thing. For anyone caught by the 4/4 anthropic lockout and trying to figure things out before the 4/17 credit deadline, this codex oauth path is honestly the cleanest workaround i've tried. [link] [comments] |
