AI代理的脑文件格式——一个文件,适用于任何大型语言模型,子毫秒级查询
每个AI代理都有健忘症。Claude不记得你上次的对话。GPT不知道你上周做出的决定。目前的解决方案——向量数据库、Markdown文件、键值存储——都失去了结构,无法追踪推理链,并且将你锁定在一个提供者上。
我构建了AgenticMemory:一种二进制图形格式,其中每个认知事件(事实、决策、推理、修正)都是一个节点,具有类型化的边(caused_by、supports、supersedes)。一个.amem文件包含你代理的整个知识图谱。可以与任何大型语言模型(LLM)配合使用。
关键数据:
• 添加一个节点耗时276纳秒
• 在一个包含10万个节点的图中,遍历5层深度耗时3.4毫秒
• 在10万个节点中进行相似性搜索耗时9毫秒
• 每年每日使用约占24MB
• 一生的记忆容量不足1GB
使用Rust构建。零依赖。Python SDK:pip install agentic-brain。Rust命令行工具:cargo install agentic-memory。
https://github.com/agentic-revolution/agentic-memory
查看原文
Every AI agent has amnesia. Claude doesn't remember your last conversation. GPT doesn't know
what you decided last week. The current solutions — vector databases, markdown files, key-value
stores — lose structure, can't track reasoning chains, and lock you to one provider.<p>I built AgenticMemory: a binary graph format where every cognitive event (facts, decisions,
inferences, corrections) is a node with typed edges (caused_by, supports, supersedes).
One .amem file holds your agent's entire knowledge graph. Works with any LLM.<p>Key numbers:
• 276ns to add a node
• 3.4ms to traverse 5 levels deep in a 100K-node graph
• 9ms similarity search across 100K nodes
• ~24 MB for a year of daily use
• A lifetime of memory fits in under 1 GB<p>Built in Rust. Zero dependencies. Python SDK: pip install agentic-brain.
Rust CLI: cargo install agentic-memory.<p>https://github.com/agentic-revolution/agentic-memory