[FEATURE] Memory Efficiency must be reviewed
Resolved 💬 3 comments Opened Mar 19, 2026 by jarir2020 Closed Mar 23, 2026
Preflight Checklist
- [x] I have searched existing requests and this feature hasn't been requested yet
- [x] This is a single feature request (not multiple features)
Problem Statement
- Sometimes claude code forgets instructions that are too old
- Sometimes claude code forgets to access global memory
- Sometimes claude code forgets to access project specific memory.
Proposed Solution
- Maybe you could use my computer's memory to save megabytes or gigabytes of memories and then add a searchtool to efficiently search the relevant instruction from the memory. This way token window is maintained and less token is wasted.
- Create a “never forget” layer which may contain references to very important instructions, global instruction locations, project specific instruction locations
- Use vector db and semantic retrival User query → embed → similarity search → inject top K memories
- Raw logs are inefficient. Use:
Summarization trees
Instruction extraction
Deduplication
- Trigger memory fetch when:
file opened
function edited
error occurs
specific keyword appears
This Reduces unnecessary retrieval
- Separate memory by scope
/global_memory
/project_A_memory
/project_B_memory
if (project == A) → only search A + global
- After producing a response: "Did I miss any relevant memory?" If yes → re-query memory → refine answer
Alternative Solutions
_No response_
Priority
Critical - Blocking my work
Feature Category
CLI commands and flags
Use Case Example
_No response_
Additional Context
_No response_
This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗