[email protected] X: @getkeyway
PEKPIK LLM Get a Key

Model access

Embedding API access for search, RAG and classification

Embedding models power semantic search, recommendation, deduplication and retrieval augmented generation. PEKPIK LLM groups embedding access with chat, code and media models so teams can manage model choice through one gateway.

Request access Read quickstart View model catalog
Primary query
Embedding API
Related searches
text embedding API / RAG embedding model API / semantic search API

Why teams search for this

Manage embedding and chat model access through one vendor surface.
Compare embedding cost separately from generation cost.
Keep model IDs configurable as your retrieval quality needs change.
Use the same procurement and billing path for RAG infrastructure and generation.

Where PEKPIK fits

Good fit

  • OKSemantic search over product docs, support content or internal knowledge bases.
  • OKRAG systems that need embedding plus generation model access.
  • OKTeams tracking embedding cost as data volume grows.

Check first

  • !Changing embedding models can require re-indexing existing vectors.
  • !Quality depends on chunking, metadata and retrieval design, not only model choice.
  • !Check vector dimensions and compatibility with your database before migration.

OpenAI-compatible example

base_url swap 
from openai import OpenAI

client = OpenAI(
    base_url="https://aiapiv2.pekpik.com/v1",
    api_key="sk-...",
)

response = client.chat.completions.create(
    model="embedding",
    messages=[{"role": "user", "content": "Summarize this for a product team."}],
)

Suggested rollout

  1. 01

    Pick a representative corpus and query set.

  2. 02

    Test retrieval quality before generating new production embeddings.

  3. 03

    Plan re-indexing if you switch models or vector dimensions.

  4. 04

    Track retrieval quality and generation quality separately.

FAQ

Can I use PEKPIK for both embeddings and chat completions?

Yes. The gateway is designed to cover several model categories, including embedding and generation models when supported.

Do I need to re-embed data when changing models?

Often yes. Different embedding models can produce different vector dimensions and similarity behavior.