AI vector embeddings of Roblox informational documents for semantic search
Grabs content from the creator-docs repo, processes and chunks them, feeds that through TogetherAI, and spits out build/
.
The GitHub workflow puts those outputs into a release so that you don't need to do this expensive step every time.
Create a .env
file:
TOGETHERAI_API_KEY=xx-XXXXXXXXXXXXXXXXXXXXXX
GITHUB_TOKEN=xxx_XXXXXXXXXXXXXXXXXXXXXXX
Then run:
pip install -r indexer/requirements.txt
python indexer/main.py
Enables fast semantic searching with vector KNN querying.
Install using wally:
[server-dependencies]
DocsAISearch = "boatbomber/robloxdocsaisearch@1.1.1"
local DocsAISearch = require(script.DocsAISearch).new({
TogetherAIKey = Secrets.TogetherAI,
GithubKey = Secrets.Github,
RelevanceThreshold = 0.3,
})
-- Optionally, preload via DocsAISearch:Load(), otherwise it'll load the index upon first query
local results = DocsAISearch:Query("how to set sun position in the sky", 2)