Auto-label your datasetin minutes — not weeks.

import os, requests
from pathlib import Path

resp = requests.post(
    "https://api.msightflow.ai/v1/label/auto",
    headers={"Authorization": f"Bearer {os.environ['MSF_API_KEY']}"},
    files={"image": Path("scene.jpg").read_bytes()},
    data={
        "tasks": "detect,segment,pose",
        "model": "yolov8m",
    },
)
result = resp.json()
print(f"{len(result['annotations'])} suggestions generated")
# Each annotation has: bbox, mask (if segment), keypoints (if pose),
# category, confidence, source='ai_generated'

# Zero-shot new class + SAM mask refinement, in one call.
resp = requests.post(
    "https://api.msightflow.ai/v1/label/auto",
    headers={"Authorization": f"Bearer {os.environ['MSF_API_KEY']}"},
    files={"image": Path("factory.jpg").read_bytes()},
    data={
        "tasks": "zero_shot,segment",
        "prompt": "cracked weld bead, spatter, undercut",   # zero-shot classes
        "segment_with": "sam",                              # refine bboxes via SAM
    },
)
# Returns: for each zero-shot detection, a SAM mask refined from the bbox

# Batch over a folder; pair with active learning to prioritise review.
import requests, glob, json

for img_path in glob.glob("dataset/*.jpg"):
    resp = requests.post(
        "https://api.msightflow.ai/v1/label/auto",
        headers={"Authorization": f"Bearer {os.environ['MSF_API_KEY']}"},
        files={"image": open(img_path, "rb")},
        data={"tasks": "detect", "project_id": "PROJECT_ID"},
    )
    # Suggestions land in project automatically

# Then sort the project queue by model uncertainty:
queue = requests.get(
    "https://api.msightflow.ai/v1/label/score-batch?project_id=PROJECT_ID",
    headers={"Authorization": f"Bearer {os.environ['MSF_API_KEY']}"},
).json()
print(f"Review these {len(queue[:50])} first (lowest-confidence)")