Unverified50% confidenceFactExact time
Context Layer会在每次发消息时自动准备相关背景知识，属于确定性触发而非等LLM决定，包含Ontology Context、Document Context和Function Backed Context三种方式
1
Sources
50%
Confidence
Long-term
Relevance
7/6/2026
First Seen
Sources
Palantir Agent与Ontology交互机制深度解析
bilibili零点未来7/4/2026
Related Claims
VerifiedContext Engineering focuses on systematically building, managing, and optimizing all information fed to a large language model, including system prompts, conversation history, external knowledge bases, and user preference profiles.75% similarUnverified上下文编译（Context Compilation）是在调用大模型之前对原始资料进行结构化处理的过程，包括扫描证据目录、生成证据清单、抽取哨兵字段、登记冲突风险四个步骤75% similarUnverifiedContextNest提供一体化整合方案，将记忆视为可组合、可嵌套的上下文单元74% similarUnverifiedContext Mode的快照机制在压缩发生前将任务状态（包括规划树、文件修改记录、决策上下文）序列化保存，压缩后再注入新的上下文73% similarUnverified/context命令可以查看当前上下文的整体大小以及不同类别分别占用的空间73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/113731API
curl https://kongchang.com/api/v1/knowledge/claims/113731MCP
get_claim(id=113731)