OpenAI-compatible embeddings API setup

What this page measured on production

On 2026-08-05, unauthenticated and invalid-key requests hit https://monkeytheme.com. No customer key and no successful vector payload were used.

Request shape (official OpenAPI)

OpenAI's public OpenAPI schema defines CreateEmbeddingRequest with these fields. Use them when the gateway is OpenAI-compatible.

Curl checks before you touch app code

# 1) Prove the route and auth layer (expect 401 JSON, not HTML)
curl -sS -D- https://monkeytheme.com/v1/embeddings \
  -H "Content-Type: application/json" \
  -d '{"model":"MODEL_ID_FROM_YOUR_LIST","input":"hello embeddings"}' | head -40

# 2) Same call with your key (redact before sharing)
curl -sS https://monkeytheme.com/v1/embeddings \
  -H "Authorization: Bearer $MONKEYAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"MODEL_ID_FROM_YOUR_LIST","input":"hello embeddings","encoding_format":"float"}'

# 3) Common path mistake: missing /v1 (production probe returned nginx 404 HTML)
curl -sS -o /tmp/no-v1.txt -w "%{http_code}\n" https://monkeytheme.com/embeddings \
  -H "Authorization: Bearer $MONKEYAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"MODEL_ID_FROM_YOUR_LIST","input":"hello"}'

Keep the HTTP status, JSON code, x-request-id / x-client-request-id headers, model id, and timestamp. Never paste a live Bearer token into tickets or screenshots.

SDK Base URL wiring

# Python
import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["MONKEYAI_API_KEY"],
    base_url="https://monkeytheme.com/v1",
)

# After GET /v1/models confirms the id:
resp = client.embeddings.create(
    model="MODEL_ID_FROM_YOUR_LIST",
    input="The food was delicious and the waiter...",
    encoding_format="float",
)
vector = resp.data[0].embedding
print(len(vector), resp.usage)

# Node.js
import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.MONKEYAI_API_KEY,
  baseURL: "https://monkeytheme.com/v1",
});

const resp = await client.embeddings.create({
  model: "MODEL_ID_FROM_YOUR_LIST",
  input: "The food was delicious and the waiter...",
  encoding_format: "float",
});
console.log(resp.data[0].embedding.length, resp.usage);

Official SDK examples call embeddings.create against the configured base URL. If Base URL already ends with /v1, do not add another /v1 in client path settings.

How to read a success response

OpenAPI defines CreateEmbeddingResponse with object: "list", model, data[], and usage. Each data[] item has object: "embedding", index, and embedding (float array when encoding_format=float).

Decision path when something fails

1. Confirm hostname monkeytheme.com and Base URL https://monkeytheme.com/v1.
2. Call GET /v1/models with the same key. If that is 401, fix auth first (401 guide).
3. Copy an exact embedding model id from the list. Do not invent names from memory.
4. POST /v1/embeddings with a short string input.
5. If you see HTML 404, the path is wrong (often missing /v1 or wrong host).
6. If you see JSON 404/model errors after auth works, treat it as model catalog or upstream support, not as “embeddings is broken for everyone.”
7. For 429 or 504 after auth works, use the dedicated rate-limit and timeout guides instead of changing the embeddings path again.

What this page does not claim

It does not claim a measured success vector, a fixed dimension count for MonkeyAI accounts, or that legacy engine paths such as /v1/engines/.../embeddings are supported. The production probe for that legacy path returned 404. It also does not replace your RAG evaluation: cosine similarity only makes sense inside one embedding model space.

Sources and related guides

• OpenAI OpenAPI schema — CreateEmbeddingRequest / CreateEmbeddingResponse field definitions used above.
• OpenAI embeddings guide — conceptual use cases and current model families.
• Why Base URL usually needs /v1
• API 401 invalid token fix
• Model not found troubleshooting
• Python SDK custom Base URL
• HTTP 429 rate limit fix

FAQ