🆘 🙉
Is this normal and nercessary to have 20k tokens usage when loaded in main session?? From system layer :
System prompt(含 Project Context):5,706 tok
Tool schemas(23 个工具的 JSON 定义):~3,516 tok
Tool summary text:大约 ~1,500 tok
Skills list:~362 tok
AND Workspace layer:
AGENTS.md:1,951 tok
SOUL.md:416 tok
TOOLS.md:213 tok
IDENTITY.md:158 tok
USER.md:120 tok
HEARTBEAT.md:42 tok
BOOTSTRAP.md:363 tok
Total: 15K tokens.🔥
Plus there's gonna be compaction, recent memory logs (summary) and MEMORY (long-term memory)...
I'm running it on Haiku4.5 as I simply burned 50USD when testing it by default setting with Sonnet 4 which wasn't affordable at all...