Unverified50% confidenceFactExact time
Agents driven by large language models do not have cross-session memory; at the start of each task, the context contains only the system prompt, the user task, and the files they are asked to read.
1
Sources
50%
Confidence
Long-term
Relevance
10/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedBecause large language models have limited context windows, the trade-off between placing information in the context and retrieving from external storage directly affects an Agent's performance on long-horizon tasks.77% similarVerifiedAgents possess capabilities for planning, tool invocation, memory, and autonomous iteration, distinguishing them from the single-round question-and-answer mode of traditional large language models.74% similarUnverifiedThe core difficulty of terminal agents lies in organizing tool calls, managing file read/write permissions, and maintaining multi-turn conversation context, rather than calling the model API.74% similarUnverifiedAn Agent's memory system is essentially a data modeling problem, not a prompt engineering problem.72% similarUnverified大模型本质是问答类型且一般没有记忆能力，而Agent不仅能问答，还能联网搜索、操作图片、识别本地资料、追踪多轮上下文72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1045935API
curl https://kongchang.com/api/v1/knowledge/claims/1045935MCP
get_claim(id=1045935)