Unverified50% confidenceFactExact time
Projects 的知识库内容在对话开始时以系统级上下文形式注入，相比每轮手动粘贴同一份文档的 token 利用效率更高，且不会随对话轮次累加重复计费
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/6/2026
First Seen
Valid until: 1/4/2027
Sources
Related Entities
Related Claims
Unverified项目级规格配置文件通过在每次会话启动时自动注入项目规范，人为构造持久化上下文，弥补LLM没有跨对话持久记忆的天然局限75% similarUnverifiedClaude Project中的知识文件会在每次对话时自动注入上下文，无需用户重复粘贴资料74% similarUnverified平台以Token消耗量而非对话轮次来衡量用量，提示词越长、上下文越多、模型越大，消耗越快73% similarUnverifiedThe Projects feature allows users to associate multiple conversation threads, upload reference files, and set Custom Instructions for contextual consistency72% similarUnverified在传统工具调用流程中，开发者通常在会话开始时就固定绑定好工具集，整个对话周期内难以变更70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/980982API
curl https://kongchang.com/api/v1/knowledge/claims/980982MCP
get_claim(id=980982)