Running long agent jobs on free DeepSeek V4.1 Flash via WorkBuddy
The author wires Hermes Agent to WorkBuddy's free DeepSeek V4.1 Flash through a local magpie gateway, runs multi-hour jobs for two days at zero credits, and shares rate-limit and quota-check lessons.
DeepSeek V4.1 Flash is free with unlimited credits on WorkBuddy AI until Oct 31, even on the free plan. So I wired it into my agent and run long jobs on my Mac from Telegram, for $0. The setup: 📱 Telegram (on my phone) → Hermes Agent on my Mac (runs scripts, browser, cron jobs) → magpie, a local gateway for all my model subscriptions → WorkBuddy AI’s free DeepSeek V4.1 Flash (1M context, tool calling) Switching Hermes over is one command: magpie hermes workbuddy-ai/deepseek-v4.1-flash Two days of multi-hour agent runs later, my credit counter still reads 0 used. Why Flash? These jobs are hours of scroll, read, click, check. Fast and free beats smart here. I keep Claude for the real thinking. Lessons from the first run: • Make it show you the list before it acts • Ask for numbers, not “done” • One browser task at a time. Running them in parallel got me rate-limited • Check the quota BEFORE each batch, not after 9 days left on the promo. What would you throw at a free 1M-context model?