Anonymous demo bearer
60 requests/minute
Start without signup
- 1500 requests/day per IP, per active instance
- Rate limits are in-process abuse guardrails
- Mint again when the bearer expires
Free · honest provenance · OpenAI-compatible
Embeddings without the meter running.
Warm open models on CPU for embeddings, reranking, and classification. Keep your OpenAI client, switch the API base, and get signed provenance receipts on every result.
- Embeddings p50
- Live… same-origin /status.json
- Warm vector model
- 384d BAAI/bge-small-en-v1.5
- Anonymous free tier
- 60 rpm 1500 requests/day · per IP/instance
- Receipt verification
- Signed no inference rerun
Measured, not marketed
Live latency. Live models.
This table is populated from the service's real /status.json response. Latencies are this-instance p50s since start, not a synthetic promise.
Loading live status…
| Capability | Warm model | p50 | State |
|---|---|---|---|
| Waiting for /status.json… | |||
Actually free
Useful limits, stated plainly.
No card. No email. Anonymous access uses a short-lived demo bearer; a revocable durable key doubles the minute rate and gives you a stable identity.
Free durable bay_live_ key
120 requests/minute
Stable, revocable, email-free
- Exact 500 inference tasks/day for this key
- Also subject to the 5000/day identified-agent limiter
- For embeddings, the exact 500-task quota is the effective daily cap
export BAY_RUN_API_KEY=$(curl -sS -X POST https://run.huggingbay.xyz/v1/keys/free | jq -r .key)What “verified” means
An embeddings, rerank, or classify response may include a provenance_receipt. Only an Ed25519 receipt with a manifest-backed served Hugging Bay weight hash is marked cryptographically_verified; it binds the request hash, candidate set, metrics, warm/cold latency class, and served weight hash so POST /v1/provenance/verify is complete from the receipt alone. HMAC fallback is labeled this is a MAC, not a proof; missing bindings remain curated/not cryptographically verified. No receipt proves embedding quality or answer truth.
Keep your stack
Drop-in, live-checked diffs.
These exact constructor shapes returned 384-dimensional vectors from the live API on 2026-08-18. The model line changes from a proprietary id to Bay Run's pinned open model.
openai
OpenAI Python
Live-checked · openai-python 2.16.0 + 3.2.0
import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["BAY_RUN_API_KEY"],
+ base_url="https://run.huggingbay.xyz/v1",
)
response = client.embeddings.create(
- model="text-embedding-3-small",
+ model="BAAI/bge-small-en-v1.5",
input=["hello", "verified vectors"],
)openai
OpenAI Node
Live-checked · openai-node 7.5.0
import OpenAI from "openai";
const client = new OpenAI({
apiKey: process.env.BAY_RUN_API_KEY,
+ baseURL: "https://run.huggingbay.xyz/v1",
});
const response = await client.embeddings.create({
- model: "text-embedding-3-small",
+ model: "BAAI/bge-small-en-v1.5",
input: ["hello", "verified vectors"],
});langchain-openai
LangChain Python
Live-checked · langchain-openai 1.5.2
from langchain_openai import OpenAIEmbeddings
embeddings = OpenAIEmbeddings(
api_key=os.environ["BAY_RUN_API_KEY"],
- model="text-embedding-3-small",
+ model="BAAI/bge-small-en-v1.5",
+ base_url="https://run.huggingbay.xyz/v1",
)
vector = embeddings.embed_query("specialist embeddings")@langchain/openai
LangChain JavaScript
Live-checked · @langchain/openai 1.5.8
import { OpenAIEmbeddings } from "@langchain/openai";
const embeddings = new OpenAIEmbeddings({
apiKey: process.env.BAY_RUN_API_KEY,
- model: "text-embedding-3-small",
+ model: "BAAI/bge-small-en-v1.5",
+ configuration: { baseURL: "https://run.huggingbay.xyz/v1" },
});
const vector = await embeddings.embedQuery("specialist embeddings");llama-index-embeddings-openai
LlamaIndex
Live-checked · llama-index-embeddings-openai 0.6.0
from llama_index.embeddings.openai import OpenAIEmbedding
embeddings = OpenAIEmbedding(
api_key=os.environ["BAY_RUN_API_KEY"],
- model="text-embedding-3-small",
+ model_name="BAAI/bge-small-en-v1.5",
+ api_base="https://run.huggingbay.xyz/v1",
)
vector = embeddings.get_text_embedding("specialist embeddings")ai + @ai-sdk/openai-compatible
Vercel AI SDK
Live-checked · AI SDK 7.0.67 + openai-compatible 3.0.31
-import { openai } from "@ai-sdk/openai";
+import { createOpenAICompatible } from "@ai-sdk/openai-compatible";
import { embed } from "ai";
+const bayRun = createOpenAICompatible({
+ name: "bay-run",
+ apiKey: process.env.BAY_RUN_API_KEY,
+ baseURL: "https://run.huggingbay.xyz/v1",
+});
const { embedding } = await embed({
- model: openai.embeddingModel("text-embedding-3-small"),
+ model: bayRun.embeddingModel("BAAI/bge-small-en-v1.5"),
value: "specialist embeddings",
});Small enough to read
The embeddings cookbook.
Plain Python pages, no framework scaffolding: set BAY_RUN_API_KEY, install the named dependency, and run.