nous portal or openrouter with a harness that uses intelligent multi provider requests,a local memory system, and pre-sub context compaction on input. if you do similar stuff often your token usage will drop after awhile of using a memory subsystem like hindsight or honcho quite a bit, and even more if you're using your harness to build relevant skills for the repeated tasks.

Do you have a harness recommendation? Sounds like maybe you’re into Hermes?