Ask HN: Which AI harness comes close to Claude Code?
I really want to try deepseek V4, but harnesss which I have previously used are inferior than Claude Code. Please suggest some Harnesses here. I have tried, roo, cline, opencode
6 comments
Why not use Claude code ? [1]
It works for me. I've read that it's not as good a match for other LLMs because it floods the context, so I keep arounf a minimal subscription to Claude itself to sort things out when deepsek goes off into the weeds, but for the most part it works well, and deepseek is a lot cheaper...
1: https://github.com/MG-Cafe/claudecode-deepseek-stack
It works for me. I've read that it's not as good a match for other LLMs because it floods the context, so I keep arounf a minimal subscription to Claude itself to sort things out when deepsek goes off into the weeds, but for the most part it works well, and deepseek is a lot cheaper...
1: https://github.com/MG-Cafe/claudecode-deepseek-stack
Half a year ago it was really hard to use any LLMs except Cloud Sonet, all others produces un-acceptable results, did a lot of changes which I did not ask to do, could not modify more then one file at a tiem, etc. Sonet was the one which really helped to develop faster. Currently almost all major providers including Deepseek passed the barrier when use of them speed up the development. I'd say, depending on the project complexity of cource, but more important becomes cost of the model. I'm trying to focus currently on the process aroud it, not model itself. If you have clearly defined phaces with acceptance criterias and check lists on each, like solution design, planning, dev, testing etc. you can even use different models for different tasks to minimize cost