Getting Started¶

Raku Sight lets you upload a labeled dataset to train your own YOLO detection model in the cloud. Once training is complete, you get a dedicated inference server with both YOLO and LLM capabilities built in — call it from any application via a simple REST API.

How it works¶

graph LR
  A[Create project] --> B[Upload dataset];
  B --> C[Train model in cloud];
  C --> D[Copy API key];
  D --> E[Call inference API];
  E --> F[Get detection results];

The platform handles dataset storage (R2), GPU training (Northflank), and versioned model deployment. You only interact with the dashboard and the API.

Step 1 — Create an account¶

Go to https://sight.raku.so/ and sign in with your operator email and password.

Step 2 — Create a new project¶

After logging in you land on the Projects dashboard. Click + New project in the top-right corner.

Give your project a name and description, then confirm. Each project has its own datasets, models, and API key.

Step 3 — Go to the Model page¶

Inside your project, navigate to the Models tab in the left sidebar.

This page lists all trained model versions for the project and their deployment status.

Step 4 — Upload a labeled dataset and start training¶

On the Models page, click Upload dataset. Select your labeled dataset archive and click Upload and start training.

Training runs on Northflank cloud infrastructure. You can monitor progress directly on the Models page — the status will update from Training → Deployed when the model is ready.

Step 5 — Copy your API key¶

Navigate to the API Keys tab in your project sidebar.

Click Copy next to your key. This key authenticates every inference request.

# Example: store as an environment variable
export RAKU_API_KEY="your-api-key-here"

Step 6 — Test or integrate¶

With your API key ready, you have two options:

Try it in the Playground — head to the Playground tab inside your project on sight.raku.so to upload images and run inference directly in the browser. No code needed — great for verifying your model is working before you ship anything.

Integrate into your app — if you're ready to call the API from your own code, see the API Integration section for the full reference and code examples in JavaScript, TypeScript, Python, .NET, Java, and Go.

API quick start¶

The base URL for all inference requests is:

https://sightapi.raku.so/api/v1

Quick example — batch prediction¶

Send one or more images and get back a complete result:

import json
import httpx

response = httpx.post(
    "https://sightapi.raku.so/api/v1/inference/predict/batch",
    data={
        "api_key": "YOUR_API_KEY",
        "query": "find me the serial number",
        "metadata": [json.dumps({"is_yolo": True, "is_llm": True})],
    },
    files=[("files", open("photo.jpg", "rb"))],
)

print(response.json())

const form = new FormData();
form.append("api_key", "YOUR_API_KEY");
form.append("query", "find me the serial number");
form.append("metadata", JSON.stringify({ is_yolo: true, is_llm: true }));
form.append("files", fileInput.files[0]);

const res = await fetch(
  "https://sightapi.raku.so/api/v1/inference/predict/batch",
  { method: "POST", body: form }
);
console.log(await res.json());

// See the Batch Prediction page for the full multipart example
result, err := predictBatch("YOUR_API_KEY", "find me the serial number",
    []string{"photo.jpg"},
    []FileMetadata{{IsYolo: true, IsLlm: true}},
)

Quick example — streaming prediction¶

Receive results incrementally as each image is processed:

import json
import httpx

with httpx.Client() as client:
    with client.stream(
        "POST",
        "https://sightapi.raku.so/api/v1/inference/predict/stream",
        data={
            "api_key": "YOUR_API_KEY",
            "query": "find me the serial number",
            "metadata": [json.dumps({"is_yolo": True, "is_llm": True})],
        },
        files=[("files", open("photo.jpg", "rb"))],
    ) as response:
        for line in response.iter_lines():
            if line:
                print(json.loads(line))

const form = new FormData();
form.append("api_key", "YOUR_API_KEY");
form.append("query", "find me the serial number");
form.append("metadata", JSON.stringify({ is_yolo: true, is_llm: true }));
form.append("files", fileInput.files[0]);

const res = await fetch(
  "https://sightapi.raku.so/api/v1/inference/predict/stream",
  { method: "POST", body: form }
);

const reader = res.body.getReader();
const decoder = new TextDecoder();
let buffer = "";

while (true) {
  const { done, value } = await reader.read();
  if (done) break;
  buffer += decoder.decode(value, { stream: true });
  const lines = buffer.split("\n");
  buffer = lines.pop();
  for (const line of lines) {
    if (line.trim()) console.log(JSON.parse(line));
  }
}

Inference modes¶

Each file you send can run one or both inference engines independently:

Next steps¶

• Batch Prediction — full request reference and code examples in all languages
• Streaming Prediction — read NDJSON chunks as results arrive
• Signed URL — generate temporary links to result assets
• API Overview — data types, error codes, and authentication reference