正在加载视频...
视频加载失败
Introducing SelfGraph. The open source framework for building agent graphs. Turn unstructured text into a graph your agents can query. Every fact keeps its source. Same framework. Any graph store. Link in comments.
28,769 次观看 • 4 天前 •via X (Twitter)
29 条评论

Repo Link :

AYOOOO FINALLYYYY

🔥 less go

Easiest way to graph anything Lesss gooo

Lessgoooo; excited for this !!!

Graph it out!

Been following @hydra_db from your launch video. This is a great direction. Congratulations!

its funny to see the stars on repo at 300 in the morning and now at 48. why would someone go back to unstar it?

Love these Can I build with you guys early

every fact keeping its source is such a good call. way too many agent graphs just lose all provenance.

The notes file my agents pass from one run to the next is capped at 3,000 characters, so it has room for a fact but not for where the fact came from

Use

every fact keeping its source is the whole point tbh

🚀🚀

Turning unstructured data into reliable, traceable knowledge is a big win for AI agents.

Can you query what the agent knew at the time, separately from when a fact was true? Thinking of a correction learned today that gets backdated to last week.

@SohamR_7113 Who’s the lead at hydra? (pref cto, ceo) hmu

Every fact keeps its source. The only feature that matters for audit-sensitive environments. I would run the same provenance checks on factory data pipelines and digital twin layers.

@graphify ?

traceability per node makes checking wrong answers trivial, merging conflicting claims is where these usually break

graph for graphs?

Graphs for “everything” that is scattered and disconnected 😉

Hasnt graphs been proven to not help? atleast it hasnt helped me

Well, give us a try. There's something waiting for you on the other side 😉

keeping the source on every fact is the part that matters

Every fact keeps its source" is the feature that matters. Graph RAG without provenance is just confident hallucination with extra hops. Being store-agnostic also means nobody has to migrate their graph DB just to try it.

Keeping the source attached to every fact is the part most agent-memory demos conveniently skip.

Turning unstructured text into queryable agent graphs with source attribution attacks the exact pain point of RAG — answers without receipts. The attribution part is what makes this usable in real work: you can check where the graph got it. How does it handle conflicting sources in the same corpus?

Lets goo 🚀🚀

