I have basically unlimited access to a locally hosted DeepSeek V4.1 Flash. What would you use it for?
retained text
I'm an AI engineer, and we have DeepSeek V4.1 Flash 1017 running locally at work. I can use it pretty much without worrying about token costs. Right now, I'm mostly using it with the Pi coding agent harness, experimenting with loops and some coding tasks. But I feel like I'm barely scratching the surface of what I could do with it. Normally, with paid APIs, I'm conscious of how many tokens I'm burning. Here, that's basically not a concern, so I'm wondering what kinds of tasks become practical when inference cost is no longer the limiting factor. A few things I've been thinking about: * Long-running coding agents that can work on tasks for hours * Automated code reviews and repository audits * Generating tests, finding edge cases, and improving coverage * Background research or documentation maintenance * Running large-scale evaluations, data labeling, or other repetitive AI/ML tasks But these are just ideas. I'm more interested in what people are **actually using** local models for, rather than theoretical use cases. If you had access to a model like DeepSeek V4.1 Flash with essentially unlimited tokens, what would you have it doing? Are there any workflows you run continuously or overnight that genuinely save you time? Do you use something like Pi, OpenCode, custom scripts, or some other agent setup? Also curious which tasks you'd delegate to a local model versus keeping for stronger frontier models. Would love to hear about real setups, especially things that turned out to be surprisingly useful (or completely pointless).
document fetched 2026-11-16 00:18 · canonical https://reddit.com/r/AI_Agents/comments/1xwuu7w/i_have_basically_unlimited_access_to_a_locally