Skip to content

Developers

Connect your software to a managed AI deployment.

weightroom operates your AI deployment. Your application sends requests through the API and uses the results in your workflow. We work with you to set up the endpoint, model, and access key.

API operations

After setup, use the base URL and service account key shown for your endpoint.

GET /v1/endpoints/{endpoint}/models

Get the model ID configured for this endpoint.

POST /v1/endpoints/{endpoint}/chat/completions

Send chat messages and receive a completion. Use the service account key as a Bearer token in the Authorization header.

Send a request

Set WEIGHTROOM_BASE_URL to the endpoint base URL shown in weightroom. Save your one-time project API key in a server-side secret store as WEIGHTROOM_API_KEY. This example uses the managed qwen3.8-27b model; for another endpoint, use its model ID from /models.

curl "$WEIGHTROOM_BASE_URL/chat/completions" \
  -H "Authorization: Bearer $WEIGHTROOM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"qwen3.8-27b","messages":[{"role":"user","content":"Summarize this note."}],"max_tokens":128}'
Open your endpoint → Connect for its quickstart

Send an image and a PDF

Install the openai Python package, set the same server-side environment variables as above, and replace receipt.png and invoice.pdf with your local files. This sends both attachments as base64 data URLs in a user message. Plain string messages still work.

import base64
import os
from pathlib import Path
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["WEIGHTROOM_API_KEY"],
    base_url=os.environ["WEIGHTROOM_BASE_URL"],
)
image_data = base64.b64encode(Path("receipt.png").read_bytes()).decode("ascii")
pdf_data = base64.b64encode(Path("invoice.pdf").read_bytes()).decode("ascii")

response = client.chat.completions.create(
    model="qwen3.8-27b",
    messages=[{
        "role": "user",
        "content": [
            {"type": "text", "text": "Compare the receipt with the invoice."},
            {
                "type": "image_url",
                "image_url": {"url": f"data:image/png;base64,{image_data}"},
            },
            {
                "type": "file",
                "file": {
                    "filename": "invoice.pdf",
                    "file_data": f"data:application/pdf;base64,{pdf_data}",
                },
            },
        ],
    }],
    max_tokens=1024,
)
print(response.choices[0].message.content)
Supported input
Text, PNG/JPEG/WebP images, and PDFs (scanned or digital, up to 10 pages). Attachments are accepted only in user messages; responses are text.
Inline files only
Remote image URLs (including HTTP and HTTPS URLs) are not accepted. Use image_url.url with an image data URL, or file.file_data with a PDF data URL and a .pdf filename. Encrypted or unreadable PDFs are rejected.
Request limits
Up to 8 image parts, up to 10 pages per PDF, and up to 12 images total after each PDF page becomes an image. Each decoded image is limited to 8 MiB and each decoded PDF to 12 MiB. The entire JSON request, including base64 data, must fit within 16 MiB.
Concurrent requests
Each endpoint serves up to 8 requests at once; additional requests wait briefly in a bounded queue. This is a configured service limit, not a measured concurrency qualification or a latency guarantee.

Managed business assistant

Qwen3.8 27B · US-MD-1

Qwen3.8 27B · 1× NVIDIA A100 SXM4, 80 GB · US-MD-1, United States. $1,899/month per endpoint, subject to current qualified capacity. Provisioning begins after verified payment. Setup and custom integrations are quoted separately.

Up to 32K context tokens · up to 4K output tokens per request. Audio, video, and tool calling are not supported. Thinking is disabled. Streaming and JSON-schema output remain candidates pending qualification; evaluate your workflow before relying on them.

Deploy a model

Model and hardware examples

These are separate assisted-setup candidates, not available managed plans. In particular, the Small L4 example is not the qualified A100 US offer above. Model and hardware fit must be confirmed with your developer.

Proposed model and hardware examples for setup estimates
PlanModel candidateHardware candidateModel/cache storage
SmallQwen3-8B1× L4, 24 GB80 GB
BalancedQwen3-14B1× RTX A6000, 48 GB100 GB
LargerQwen3-32B1× A100, 80 GB160 GB
Discuss other model configurations

Separate product

GPU Workspaces

Rent a self-service GPU environment for your own tools, development, or experimentation. You manage the software and setup inside the workspace; this is separate from an operated private AI deployment.

Explore GPU Workspaces

Need integration or setup help?

Tell us about the task and the software you want to connect.

Get setup help