what other alternative harness options do i have (small setup)
retained text
My goals is to use a harness that isn't too complex, isn't too demanding on my hardware, and deploy agents at a given time to complete various of task Task include: gmail reading, searching information from the web, creating/moving folders i really love **agent loops** \~I'm not sure if its realistic but the **harness** could probably swap models to do different task, for example my qwen3.6 35b could do the main task. When its done, swap to a smaller model with vison to describe the output (screen shot then describe). Once done, it swaps back to the main model to continue\~ My setup: \-rx6700xt 12gb vram \-r5 5600x \-32gb ddr4 3200 Models: \-qwen3.6 35b a3b **MTP** q4\_km unsloth, --spec-type draft-mtp can't pair with mmproj in llama.cpp \-qwen3.8 27b iq3\_xxs unsloth thing's ive tried Deepseek harness, Hermes Agent, Pi, Opencode, n8n, openwebui, openclaw For me, **Hermes Agent** has everything i need, from coding, agent loop etc. But its extremely heavy on my setup and im looking for alternatives **n8n** is extremely complex + difficulty setting things up and many errors **Pi/Opencode** if im not mistaken, are only good for coding **Openclaw** kinda fell off for **Deepseek Harness**, im still experimenting with it since i have to manually add skills like openwebui. I’ve already played with openwebui for a few months now and didnt make any progression, I’m looking for alternatives Other than that im looking for alternatives harness that could potential check out these goals, i appreciate any help thanks
document fetched 2026-09-15 00:43 · canonical https://reddit.com/r/LocalLLaMA/comments/1wg1hd4/what_other_alternative_harness_options_do_i_have