Voyage AI

Last updated: Sep 22, 2026

---

Rationale

Voyage AI is our provider for embedding models.

The main reasons why we chose it over other alternatives are the following:

• It is highly accurate.
• It has low latency.
• It is the only one with a model specific to code.
• It offers easy consumption through API endpoints.
• The pricing model is based on usage.

Alternatives

The following alternative was considered but not chosen for the specified reason:

OpenAI Vector Embeddings: It does not offer code specific embedding.

Usage

We use Voyage AI for code embeddings in two components:

• Matches embeds the titles and descriptions of weaknesses with voyage-3 and runs a cosine nearest-neighbor search over the vectors to match vulnerabilities.
• Sifts embeds function snippets with voyage-code-3 and feeds the vectors to a local candidate classifier. Sifts retrieval is agentic and does not use vector search.