deadbits / vigil-llm

⚡ Vigil ⚡ Detect prompt injections, jailbreaks, and other potentially risky Large Language Model (LLM) inputs
https://vigil.deadbits.ai/
Apache License 2.0
277 stars 33 forks source link

Add new embeddings via API #4

Closed deadbits closed 10 months ago

deadbits commented 10 months ago

Users will want to add their own embeddings (for custom detections, new techniques, etc). The API and command line should allow easy indexing to the database.

This should be really simple.. just init and call the exsiting embedding.VectorDB.add_texts(). But then there will be two chroma clients open when the vectordb scanner is enabled. Idk if this is really even an issue but something to keep in mind.

deadbits commented 10 months ago

Now that the vectordb is decoupled from the scanner, this is as simple as adding a new endpoint that calls add_texts. Server can parse config and create a vectordb thats used for this endpoint and the scanner

deadbits commented 10 months ago

Done via https://github.com/deadbits/vigil-llm/releases/tag/v0.6.2-alpha