Ability to capture cache locally and re-use later.
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
I come back to a session after a while and the first prompt when I resume consumes a large amount of my usage.
This is for when I want to maintain the full session without /clear or /compact.
Proposed Solution
First, I feel it is reasonable for Anthropic to remove a cache after a certain amount of time...memory is money and there is high demand.
My solution is for when I do want to resume a session in the future, have the ability to /download-cache to my local computer. Then, when I resume, it gets uploaded into Anthropic memory without requiring a big conversation prompt.
The hit on the user's end is some bandwidth usage and time. The hit for Anthropic is to figure out how to serialize+unserialize a cache block.
The benefit to the user is to drastically reduce that initial prompt hit on usage.
Alternative Solutions
/clear and /compact are generally useful to manage token usage...but with 1M context window, many times it's nice to keep a longer conversation going or have lots of code in context.
Priority
Medium - Would be very helpful
Feature Category
API and model interactions
Use Case Example
- I'm working on a project and I have my configurations, claude.md, and a bunch of code in context.
- I leave the session for a couple of hours.
- Currently: I come back and my first chat consumes 25% of my 5-hour usage.
- Proposed: before I leave I have a command that downloads the cache to my computer so when I come back the usage hit is lessened.
Additional Context
_No response_