Free embeddings API. No key, no account.
OpenAI-compatible text embeddings powered by NVIDIA Nemotron Embed 1B. Send text, get back a 2048-dimensional vector. No sign-up, no API key, no credit card — the endpoint is free to call.
Quickstart
One POST request. Works with the OpenAI SDK out of the box.
cURL
curl https://polygon.mapleai.shop/v1/embeddings \
-H 'content-type: application/json' \
-d '{
"input": "MapleAI sells GPT access per request",
"input_type": "query",
"encoding_format": "float"
}'
JavaScript (OpenAI SDK)
import OpenAI from 'openai';
const client = new OpenAI({
baseURL: 'https://polygon.mapleai.shop/v1',
apiKey: 'not-required'
});
const res = await client.embeddings.create({
model: 'nvidia/nemotron-3-embed-1b',
input: 'MapleAI sells GPT access per request'
});
console.log(res.data[0].embedding.length); // 2048
Python (OpenAI SDK)
from openai import OpenAI
client = OpenAI(
base_url="https://polygon.mapleai.shop/v1",
api_key="not-required"
)
res = client.embeddings.create(
model="nvidia/nemotron-3-embed-1b",
input="MapleAI sells GPT access per request"
)
print(len(res.data[0].embedding)) # 2048
Batch up to 128 inputs
curl https://polygon.mapleai.shop/v1/embeddings \
-H 'content-type: application/json' \
-d '{
"input": ["first document", "second document"],
"input_type": "passage"
}'
Use input_type: "query" for search questions and
"passage" for documents — asymmetric retrieval works best
when the two sides are marked differently.
Reference
Request and response format.
| Field | Type | Description |
|---|---|---|
input (required) | string | string[] | One text or an array of 1–128 texts |
model | string | Optional; any value is accepted and routed to nvidia/nemotron-3-embed-1b |
input_type | query | passage | Retrieval role of the text |
encoding_format | float | base64 | Vector encoding, default float |
Response
{
"object": "list",
"data": [
{
"object": "embedding",
"index": 0,
"embedding": [0.0123, -0.0456, ...]
}
],
"model": "nvidia/nemotron-3-embed-1b"
}
Errors carry the schema
Invalid requests return HTTP 400 with a machine-readable
details.schema describing the exact valid shape, so an
agent can self-correct on the first try.
Use cases
What people build with free embeddings.
Semantic search
Embed your document corpus once, embed the query at search time, rank by cosine similarity. The query/passage input types are built for exactly this.
RAG pipelines
Retrieve relevant chunks with free embeddings, then answer with a paid GPT model — you only pay for the final generation. See the runnable embeddings-to-chat example.
Clustering and dedup
Group near-duplicate tickets, reviews or crawler output before sending anything to a paid model. 2048-dim float vectors work with FAISS, pgvector, Qdrant and friends.
Fair use
Free means free, with common-sense limits.
What is included
Unauthenticated access for individuals, agents and prototypes. No API key, no account, no payment. The model is fixed to nvidia/nemotron-3-embed-1b.
What is not
Bulk corpus re-embedding at datacenter scale. If you push sustained high volume, we may rate-limit by source IP. Need guaranteed throughput? Talk to us: [email protected].
Next step
Found something? Answer with GPT, pay per request in USDC.
GPT via x402
4 GPT models behind the same OpenAI-compatible shape, paid per request in USDC on Solana, Base, Polygon or Arc. No account — your wallet is your identity. See models and prices.
Prepaid API keys
Prefer a budget? Buy a token pack for one GPT model (0.1M–1M tokens from $0.007) and call mapleai.shop/v1 with a bearer key. Prepaid guide.