Why?
- session management - accurate (unlike OpenCode last time I tried), tree structured (allows me to jump back and forth to manage context growth), compaction is just another message and pi manages which subtree to send to the model
- fast
- super customizable
- no hidden stuff
- sane defaults - it is a shame that this is a differentiator.
PI is best for any local model not just Qwen but you can use hermes as it offers more kind of productivity though hermes is not optimized right for local agents which will somewhat affect your work balance but with some tweaks like
1.Trim the Token Tax: Run hermes tools and turn off everthing unused
2.Enable KV Cache / Prefix Caching
I can’t tolerate anything other than Pi anymore.
Why? - session management - accurate (unlike OpenCode last time I tried), tree structured (allows me to jump back and forth to manage context growth), compaction is just another message and pi manages which subtree to send to the model - fast - super customizable - no hidden stuff - sane defaults - it is a shame that this is a differentiator.
PI is best for any local model not just Qwen but you can use hermes as it offers more kind of productivity though hermes is not optimized right for local agents which will somewhat affect your work balance but with some tweaks like
1.Trim the Token Tax: Run hermes tools and turn off everthing unused 2.Enable KV Cache / Prefix Caching
I've been mostly drawn to pi simply because I can modify it so easily to have extensions or behavior I want.
pi.dev with
"packages": [ "npm:context-mode", "npm:pi-subagents", "npm:pi-web-access", "npm:pi-bar", "npm:pi-opencode-bridge", "npm:@ssweens/pi-leash", "npm:pi-caveman", "npm:@ff-labs/pi-fff", "npm:pi-ui-hephaestus", "npm:@alexanderfortin/pi-zai-usage", "npm:pi-cache-optimizer", "npm:@hypabolic/pi-hypa" ]
which extension plugin do you use to debug/automate the web browser?
you shall use CC, serve it via ollama, you need to patch it as CC's timestamp ruins prefill cache.
You need a harness? Does llama include one?
llama.cpp has a web ui nowdays, but with an harness is easier to interact with existing cli tools
It must have some kind of harness at least?? Like “you’re an AI assistant, answer questions”?
opencode, pretty good setup out of the box, no need to go through and find packages for the basics like pi
there are MCP/skills for browser integration that work in any harness