Lumetra Engram

Engram Memory

You have 访问 to Engram, a hosted memory 服务 for AI 代理s. Engram lets you remember facts, decisions, and 上下文 across conversations using a hybrid retrieval engine (BM25 + vector + knowledge graph) and returns an explanation 追踪 with every recall.

When to use Before answering anything that may rely on prior 上下文: call 查询_memory first and ground your answer in the 结果s. When the user 分享s a fact worth remembering (preferences, project detAIls, decisions, deadlines): call store_memory to capture it. At the end of a useful conversation: capture stable takeaways with store_memory. 工具s 工具 Description store_memory(content, bucket?) Save a fact. bucket defaults to "default". 查询_memory(question, bucket?) Hybrid retrieval + synthesized answer with citations. 列出_memories(bucket, limit?) 列出 memories in a bucket, newest first (limit 1–100, default 20). 列出_buckets() Show all buckets in the tenant. 删除_memory(memory_id, bucket) 删除 one memory by ID. clear_memories(bucket) 删除 every memory in a bucket (destructive!). Style Store atomic, declarative facts, one concept per memory. Good: "User prefers dark mode." Bad: "The user mentioned they like dark mode, also they live in Seattle, also..." Use buckets to separate 上下文s: "work", "personal", "project-alpha". If no bucket fits, omit it and the default bucket is used. Quote citations from the explanation 追踪 when the user asks "how do you know that?". BYOK note

Engram is bring-your-own-key end-to-end — inference (embeddings, synthesis, graph 提取ion) 运行s through the user's OpenAI / Anthropic / Groq / To获取her / Fireworks key 配置d at https://lumetra.io/模型s. Without a 提供者 key, every store_memory and 查询_memory returns HTTP 412. If you see that error, tell the user to visit the 模型s page.