Unverified50% confidenceDecision RuleExact time
In NoteBank's addressing stage, each note stores a compact address for each layer-head channel (the native query vector at the note's most unique token); when a question arrives, each channel independently computes cosine similarity and votes, and the top R notes with the most votes are retrieved, with earlier notes preferred in case of ties.
1
Sources
50%
Confidence
Long-term
Relevance
10/15/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedKey constraints during memory storage include checking for duplicates before writing, prioritizing updates to existing files, and keeping Descriptions accurate.61% similarUnverifiedIn a memory test where "account ID 7731, net-45 terms" was deliberately planted, the full history remembered it 3/3 times, the most recent 6-turn sliding window remembered it 0/3 times (answering "account ID not received"), and summary + most recent 4 turns remembered it 3/3 times.60% similarUnverifiedA text note system maintained by one developer has about 220 notes and nearly a million tokens, with one note per topic, the current state at the top and a dated history at the bottom.60% similarUnverifiedIn multi-turn interactions, each round's input must carry the complete conversation history (context window), causing the number of input tokens in the Nth turn to be approximately equal to the sum of the content from the preceding N-1 turns.60% similarUnverifiedPrompt caching reuses the already-computed intermediate state of the same long context submitted repeatedly (such as a fixed large document or system prompt), thereby reducing the billing cost for that portion of the input.59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1040234API
curl https://kongchang.com/api/v1/knowledge/claims/1040234MCP
get_claim(id=1040234)