Labelbox

Labelbox is enterprise-grade data labeling infrastructure — widely used for building training datasets for computer vision, NLP, and multimodal ML. For agents in ML pipelines: programmatically import data, create labeling projects, retrieve annotations, and export labeled datasets for model training. Model-assisted labeling (MAL) enables agents to pre-label batches and route uncertain samples to human review. Strong GraphQL API gives flexible query patterns. Confidence is docs-derived.

REST and GraphQL API at api.labelbox.com/graphql. Python SDK (pip install labelbox) wraps the API. Core resources: projects, datasets, data_rows, labels, model_runs. lb.Project.create(name) creates a labeling project. lb.Dataset.create(name).create_data_row(row_data=url) adds items for labeling. Export: project.export_v2() retrieves all labels in NDJson format. MAL: model.upsert_predictions(data_rows) submits pre-labels. GraphQL: query project { dataRows { labels { jsonAnswer } } } for flexible label retrieval.

API key auth: lb.Client(api_key=...) or LABELBOX_API_KEY env variable. Keys from Labelbox Account settings. HTTPS enforced. No OAuth2. Dataset and project access controlled by workspace membership. Enterprise: SAML SSO available.

Python SDK raises lb.exceptions.LabelboxError subclasses for API errors. Rate limiting: per-account limits (generous for ML batch operations). Export is async — large exports use job-based pattern with status polling. Labelbox uptime at status.labelbox.com. Data storage (assets) backed by cloud storage (S3/GCS); data row accessibility depends on asset URL validity.

docs.labelbox.com covers Python SDK guide, GraphQL reference, project setup, annotation types, MAL workflow, and export format documentation. Getting started: sign up (free tier available), install SDK, create first project in under 30 minutes. Strong ML-oriented documentation. Enterprise support for large-scale labeling operations. Community via Labelbox Slack.