Get Started with Open Models & the OpenAI SDK

1. Get your key and model ID

Create an account or sign in to the console.
In Get started, create an API key with inference access. Save it securely: the full key is shown only once. Manage existing keys from your account’s Keys page.
Open Models, choose a chat-capable model. Copy its exact model ID from View code.

Base URL: https://api.models.blade.sh/v1
Authentication: Authorization: Bearer YOUR_BLADE_API_KEY
First endpoint: POST /v1/chat/completions

Use your Blade key with this URL; an OpenAI key will not authenticate here.

2. Set up your project

Python

Use Python 3.10 or newer with pip. In a new project directory, create a virtual environment and install the OpenAI SDK. These commands use a macOS or Linux shell.

Terminal · Python setup

python3 -m venv .venv
. .venv/bin/activate
python3 -m pip install openai

Windows / PowerShell setup

PowerShell · Python setup

py -m venv .venv
.\.venv\Scripts\python.exe -m pip install openai

Use .\.venv\Scripts\python.exe hello.py to run the example without activating the environment.

JavaScript / Node.js

Use Node.js 22 or a newer supported LTS release with npm. Run this in a new project directory. In an existing npm project, only the install command is needed.

Terminal · Node.js setup

npm init -y
npm install openai

The examples run on your server with Node.js. The .mjs extension enables JavaScript module imports.

curl

Use a recent version of curl that supports --fail-with-body. The commands below use a macOS or Linux shell; no SDK is required.

Set the following environment variables in the terminal you will use to run the example. Replace the placeholders with your key and model ID.

Terminal · Environment

export BLADE_API_KEY="YOUR_BLADE_API_KEY"
export BLADE_MODEL="YOUR_AVAILABLE_MODEL_ID"

Windows / PowerShell environment variables

PowerShell · Environment

$env:BLADE_API_KEY="YOUR_BLADE_API_KEY"
$env:BLADE_MODEL="YOUR_AVAILABLE_MODEL_ID"

BLADE_API_KEY and BLADE_MODEL are local names used by this guide. The examples read them explicitly; creating a .env file alone will not load them.

Terminal · Environment

export BLADE_API_KEY="YOUR_BLADE_API_KEY"

You will put the model ID in request.json in the next step.

3. Send your first request

Python

Save this file as hello.py in your project directory.

hello.py

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.models.blade.sh/v1",
    api_key=os.environ["BLADE_API_KEY"],
)

response = client.chat.completions.create(
    model=os.environ["BLADE_MODEL"],
    messages=[{"role": "user", "content": "Say hello in one sentence."}],
)
print(response.choices[0].message.content)

Terminal · Run the request

python3 hello.py

JavaScript / Node.js

Save this file as hello.mjs in your project directory.

hello.mjs

import OpenAI from "openai";

const { BLADE_API_KEY, BLADE_MODEL } = process.env;
if (!BLADE_API_KEY || !BLADE_MODEL) {
  throw new Error("Set BLADE_API_KEY and BLADE_MODEL first.");
}

const client = new OpenAI({
  baseURL: "https://api.models.blade.sh/v1",
  apiKey: BLADE_API_KEY,
});

const response = await client.chat.completions.create({
  model: BLADE_MODEL,
  messages: [{ role: "user", content: "Say hello in one sentence." }],
});
console.log(response.choices[0].message.content);

Terminal · Run the request

node hello.mjs

curl

Save this JSON as request.json. Replace YOUR_AVAILABLE_MODEL_ID with the model ID you copied from the console.

request.json

{
  "model": "YOUR_AVAILABLE_MODEL_ID",
  "messages": [
    { "role": "user", "content": "Say hello in one sentence." }
  ]
}

Run this command from the same directory as request.json.

Terminal · Send the request

curl --fail-with-body --silent --show-error \
  https://api.models.blade.sh/v1/chat/completions \
  -H "Authorization: Bearer $BLADE_API_KEY" \
  -H "Content-Type: application/json" \
  --data-binary @request.json

Read the result

The SDK examples print the model’s answer. curl returns the full JSON response; the answer is in choices[0].message.content. The wording varies between runs.

choices[0].message.content: The generated answer.
choices[0].finish_reason: Why generation ended. A value of length means the output reached a limit; review the token budget.
usage: Token counts returned with the response. These are not a currency amount; see how inference pricing works.

If the request fails, use the status code and error message to identify the cause.

4. Stream the answer

Set stream to true to receive the output as it is generated. Check that the selected model supports streaming.

Python

Save this complete example as stream.py and run python3 stream.py in the same environment. On Windows, use .\.venv\Scripts\python.exe stream.py.

stream.py

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.models.blade.sh/v1",
    api_key=os.environ["BLADE_API_KEY"],
)

stream = client.chat.completions.create(
    model=os.environ["BLADE_MODEL"],
    messages=[{"role": "user", "content": "Say hello in one sentence."}],
    stream=True,
)
for chunk in stream:
    if chunk.choices:
        print(chunk.choices[0].delta.content or "", end="", flush=True)
print()

JavaScript / Node.js

Save this complete example as stream.mjs and run node stream.mjs in the same terminal.

stream.mjs

import OpenAI from "openai";

const { BLADE_API_KEY, BLADE_MODEL } = process.env;
if (!BLADE_API_KEY || !BLADE_MODEL) {
  throw new Error("Set BLADE_API_KEY and BLADE_MODEL first.");
}

const client = new OpenAI({
  baseURL: "https://api.models.blade.sh/v1",
  apiKey: BLADE_API_KEY,
});

const stream = await client.chat.completions.create({
  model: BLADE_MODEL,
  messages: [{ role: "user", content: "Say hello in one sentence." }],
  stream: true,
});
for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta.content ?? "");
}
process.stdout.write("\n");

curl

Save this as stream.json and replace the model placeholder. --no-buffer lets curl display incoming events immediately.

stream.json

{
  "model": "YOUR_AVAILABLE_MODEL_ID",
  "stream": true,
  "messages": [
    { "role": "user", "content": "Say hello in one sentence." }
  ]
}

Terminal · Stream the request

curl --no-buffer --fail-with-body --silent --show-error \
  https://api.models.blade.sh/v1/chat/completions \
  -H "Authorization: Bearer $BLADE_API_KEY" \
  -H "Content-Type: application/json" \
  --data-binary @stream.json

curl displays raw server-sent events: data: lines containing JSON chunks, followed by data: [DONE]. Your app should parse these events instead of displaying the raw stream.

Read text from choices[0].delta.content. Some chunks carry no text or have an empty choices array; the SDK examples handle both. For streaming token totals, set stream_options.include_usage to true and read usage from the final usage chunk.

Check support before adding features

OpenAI-compatible means you can reuse the client for supported routes. It does not mean every OpenAI feature or parameter is available.

Feature	What to check
Chat Completions	Use `/v1/chat/completions`. Check each model’s capabilities, context window and `supported_parameters` in the model reference or `GET /v1/models`.
Conversation history	Send the relevant earlier messages with each request. The client does not preserve your conversation automatically.
Responses	The current `/v1/responses` API is stateless. `store: true` and `previous_response_id` are not supported. Replay history in the request instead.
Function calling	Choose a model with native tool calling. Your application executes client-defined functions and returns their results.
Staan web search	Enable web search for the account and choose a model with native tool calling. Search is opt-in per request and billed per executed search. Search queries go to Staan; see data handling and the API reference before enabling it.
Other server tools	Tool access and model support are required. Check the console’s Tools page for activation and billing details.

List models from your terminal

With a key that includes models:read, this read-only request returns a JSON list in data. Inspect the model’s id and supported parameters before using it.

Terminal · List models

curl --fail-with-body --silent --show-error \
  https://api.models.blade.sh/v1/models \
  -H "Authorization: Bearer $BLADE_API_KEY"

When a request fails

Read the error body as well as the HTTP status. Retrying an invalid key, unsupported parameter or exhausted daily quota will not fix the request.

Status	What to do
`400` · Request	Check JSON syntax, accepted parameters, the model’s output-token cap and context limit. Remove or correct the field named in the error.
`401` · Key	Confirm the environment variable is set in the same terminal. Check for a missing, malformed, expired or revoked key and use a Blade key with the Blade base URL.
`403` · Access	Check key permissions and the account’s model, external-provider or tool access. Ask an account administrator to enable the required access.
`404` · Model / route	Copy the exact model ID from the console and check the route. Set the SDK base URL to `https://api.models.blade.sh/v1`; do not append `/chat/completions` to the SDK base URL.
`413` · Request size	Reduce the request body to fit the endpoint’s documented size limit.
`429` · Limits	For `rate_limit_exceeded`, reduce concurrency and respect `Retry-After`. For `insufficient_quota`, check the daily token allowance; the documented reset is midnight UTC.
`502–504` · Upstream	Use a limited number of retries with a delay for transient failures; investigate repeated failures rather than retrying indefinitely.

The example fails before making a request

Python cannot import openai: use the same interpreter for installation and execution. Activate .venv, or run its Python executable directly.
Node.js cannot find openai: install it in the project directory and keep the example’s .mjs extension.
A variable is missing: set it in the current terminal, then run the example there. A new terminal will not inherit variables exported in another session.
curl cannot read the file: run the command in the directory containing request.json or stream.json.

Check usage. Then build on it.

Open Usage in the console and select the relevant date range. Review requests and token counts by model and key, and export CSV when needed. Displayed costs are estimates; the invoice is authoritative.

For audio, image generation, OCR or embeddings, start from the corresponding console studio and the model’s API example. These workloads have their own inputs and routes; a text-chat request is not a universal template.

PricingUnderstand the cost of a request →ReferenceExplore routes and parameters →

Checked on 16 September 2026 against the console and Blade’s quickstart and API reference. SDK syntax: official Python and JavaScript references. Model availability and enabled features depend on your account.

Your SDK.Your first model call.

1. Get your key and model ID

2. Set up your project

Python

JavaScript / Node.js

curl

3. Send your first request

Python

JavaScript / Node.js

curl

Read the result

4. Stream the answer

Python

JavaScript / Node.js

curl

Check support before adding features

When a request fails

Check usage. Then build on it.

Your SDK.
Your first model call.