What TGM is

Temporal Graph Memory (TGM) turns your documents into a knowledge graph and answers questions with exact passages from them. There is no generative language model in the retrieval engine; returned passages come from your documents.

How it works

1. You send documents through the API.
2. TGM splits them into passages and connects the passages into a graph, using embeddings alone.
3. You ask a question in plain language.
4. You get the exact passages back, each naming the file it came from. If the answer is not in your documents, you get not_covered: true instead of a guess.

What makes it different

TGM serves passages from your own files or returns a coverage signal. A passage matching its source does not, by itself, establish that it completely answers your question. An answer is always checkable: every passage names its source, and asking the same question again returns the same answer.

Every change to a graph is also a saved version. You can ask last month’s version a question and get last month’s exact answer, which makes answers auditable long after the documents change.