Requesting comment on a soon-to-be-published article

[JBailes]

Hello,

The repo readme states to file an issue for any questions.

I am planning to publish an article around this project in the next 24 hours, and there is a journalistic right-to-reply for these cases.

You can review the pre-publish draft at https://github.com/RakuenSoftware/rakuen-blog/blob/main/articles/your-memory-has-no-authority-model/article/your-memory-has-no-authority-model.md

Audited commit: c406932

1. Our reading is that edge_type_map supplies type definitions to the extraction prompt, but the write path does not reject an extracted relationship for violating those types. Is that accurate? If not, where is the runtime validation performed?
2. Does an EntityEdge or related record contain a structured authority field distinguishing a direct user assertion from a model inference? Can that distinction prevent an inferred claim from expiring or superseding a user assertion?
3. Can every retrieved fact be traced to the exact originating episode? Is there also a first-class document or source object above the episode?
4. Does the open-source implementation support source-code structure and ordinary organizational memory in the same graph and retrieval operation, or is that outside Graphiti’s intended scope?

[danielchalef]

Hello - Graphiti is not an agent memory service. It is a framework for building temporal knowledge graphs. Your article should be explicit that this is the case.

Building graphs from source code is out of scope for Graphiti. We'd encourage you to include Graphify in your article if this is the intended focus: https://github.com/Graphify-Labs/graphify

Regarding your questions: Most all of these are memory features available in Zep, but are not available in Graphiti.

Daniel