[Concept] : PG Vector - Extension / Proof of Concept

[h0lybyte]

Core Concept/Theory
A clear and concise description of what the concept is. Ex. It would be cool if [...]

We want to create an isolated issue ticket to keep track of the initialization of pgvector and a simple memory store.
For this, we have to adjust the docker image of kilobase to include the pgvector extension.
Afterwards, we need to create a set of default memory stores via PSQL.
Finally, create a couple test integrations via Discord / IRC and see if we can utilize a message history in the LLM conversations.

This ticket is just to encompass all of these tasks into a central location, for referencing and tracking.

---

Alternative Ideas
Is there any other way this concept could be used?

Not right now, I can always loop back around and add more here as a reference.

---

Alternative Examples/Sources
Are there any other references that you can provide?

I will include alternative examples and sources later on.

---

Additional information
Add any other context or examples of this concept here.

No additional information as of right now.

[h0lybyte]

TODO:

[1] Bump the version of supabase to 15.8.x within the kilobase docker image.
[2] That also means bumping up the studio container to 2025~ release.

From

https://github.com/supabase/supabase/blob/master/docker/docker-compose.yml

[h0lybyte]

VectorChord Extension Added to KBVE/postgres

PR: KBVE/postgres#10

What was added:

• VectorChord (vchord) 1.1.0 — scalable, disk-friendly vector similarity search extension for PG17
• Bootstrapped pgrx 0.17.0 with Rust 1.90.0 (new toolchain required by VectorChord)
• Created reusable tools/pgrx-hash/ utility for bootstrapping future pgrx versions via Docker
• CI smoke tests now verify vector, kilobase, and vchord all load correctly

How it relates to this ticket:
VectorChord builds on top of pgvector (CREATE EXTENSION vector is a prerequisite) and provides high-performance IVF/HNSW indexing optimized for disk-based workloads — directly supporting the memory store use case outlined here.

[h0lybyte]

Testing the smoke test for the extension but once everything passes, we should be good to go!

Going to close this out because it's no longer a direct extension issue but rather just a nix problem.