docs: clarify cache-ttl pruning window
This commit is contained in:
@@ -13,6 +13,7 @@ Session pruning trims **old tool results** from the in-memory context right befo
|
||||
- Only affects the messages sent to the model for that request.
|
||||
- Only active for Anthropic API calls (and OpenRouter Anthropic models).
|
||||
- For best results, match `ttl` to your model `cacheControlTtl`.
|
||||
- After a prune, the TTL window resets so subsequent requests keep cache until `ttl` expires again.
|
||||
|
||||
## What can be pruned
|
||||
- Only `toolResult` messages.
|
||||
|
||||
Reference in New Issue
Block a user