Expired70% confidenceFactExact time
An agent's memory includes short-term memory (conversational context window, typically 4K–128K tokens), long-term memory (vector database), and procedural memory (intermediate state cache for tool calls).
4
Sources
70%
Confidence
Medium-term (~90 days)
Relevance
7/21/2026
First Seen
Valid until: 10/19/2026(expired)
Sources
Related Claims
UnverifiedShort-term working memory, i.e., the context window, is typically 8K to 128K tokens, while long-term memory is usually stored in the form of a vector database82% similarUnverifiedThe conversational memory mechanism is essentially concatenating historical messages into the context window, so context length limits (e.g., 4K, 8K, 128K tokens) become a key constraint.79% similarUnverifiedMemGAS在LongMemEval-S上构建记忆处理约52.9M input tokens，Summary和Keyword带来的额外内存约27MB，约为原始记忆的10%78% similarUnverifiedThe memory system in the Harness layer typically adopts a multi-level architecture: short-term memory uses a sliding-window conversation history, medium-term memory uses summarization compression techniques, and long-term memory uses vector databases (such as Pinecone, Milvus, and Chroma) for persistent storage.76% similarUnverified智能体对内存要求较高，64GB 内存可秒出结果，32GB 或 16GB 首次对话会较慢需要预热过程75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/579861API
curl https://kongchang.com/api/v1/knowledge/claims/579861MCP
get_claim(id=579861)