Repository navigation
Exploring Signet Cloud for background inference #647
NicholaiVogel
announced in
Announcements
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Signet works best when it can think in the background.
A longstanding pain-point in Signet is that the memory pipeline requires continuous inference for surfacing insights through background distillation, drawing meaningful connections in the knowledge graph, and continuously updating it. That means users are either forced to:
A. use their existing subscription-based AI plans, by letting Signet call their preferred CLI headlessly. (we just added ACPX as an official inference backend, still in beta, but this will be getting cleaner and more reliable for users that prefer to operate this way)
B. run inference locally using llama.cpp, vllm or ollama
C. turn off background processing/inference entirely.
So while users get to own their memory, most are not financially able to support the level of inference or the compute necessary to make the most of it. And that sucks so much.
So what I'm thinking is that we could build a simple cloud/API service that helps take care of the baseline inference needs for Signet users. We are exploring whether we can provide a small free monthly allowance for background inference, so new users can experience Signet properly without configuring a model or API key on day one. That could include multi-modal embeddings and a set number of tokens per month for the background pipeline, managed by Signet Cloud.
A level up from this could be a paid subscription service, ideally for around $4-8/month, that would offer more tokens, hosted sync, and best-in-class embedding models.
All reactions