Bespoke Curator

What it does

Bespoke Curator is an open-source Python library for building synthetic data pipelines. You feed it rows of input data, such as a list of reviews or a set of documents, along with a prompt and a structured output schema. It sends those requests to language model providers and returns a dataset of parsed results, which can be loaded into a Hugging Face Dataset object.

People use it to create reasoning and training datasets, run sentiment or feature extraction over large collections, and generate fine-tuning data. Its distinguishing points are structured outputs as a first-class feature, built-in handling for parallelism, retries and caching, and batch mode, which the page says can halve token costs on supported APIs. It also offers code execution through several backends and fine-tuning integrations with Tinker and Fireworks AI.

Curator suits developers and ML practitioners comfortable with Python who need to process large volumes of model calls. It requires installing the bespokelabs-curator package and setting provider API keys as environment variables. Model costs are paid to the underlying provider. Fireworks AI is an optional add-on, and without its SDK or an API key the trainer runs in mock mode.

What it can do

• Generate synthetic data at scale through a Python class that wraps model calls
• Enforce structured outputs with Pydantic models
• Batch processing with OpenAI and Anthropic batch APIs, set with batch=True
• Built-in parallelism, retries, and response caching
• Code execution through local multiprocessing, Ray, Docker, or e2b backends
• Managed fine-tuning integrations with Tinker and Fireworks AI

Works with

• OpenAI
• Anthropic
• LiteLLM
• vLLM
• Gemini
• Hugging Face Datasets
• Tinker
• Fireworks AI
• Together.ai

Before you choose

• Model API usage is billed by the underlying provider; the page does not state Curator's own rates.
• Shared bespokelabs namespace with Sandbox and Nimble packages can cause import conflicts if older editable installs are present.
• Fireworks AI integration is an optional add-on and runs in mock mode without its SDK or API key.

Listed under

• Data

Get started

Run pip install bespokelabs-curator, then set your model provider API keys.

1. Install the package with pip install bespokelabs-curator, using a fresh virtual environment if upgrading.
2. Set the API key for your model provider as an environment variable, such as OPENAI_API_KEY or ANTHROPIC_API_KEY.
3. Define a subclass of curator.LLM with prompt and parse methods, then run it on a list or Hugging Face Dataset.

The open-source library is licensed Apache-2.0. Model and batch API costs are paid to providers. Hosted console pricing was not confirmed on the source page.

We couldn’t find published prices for Bespoke Curator. See their pricing page ↗

Sources for this overview

Product pages, documentation, and pricing checked 30 Oct 2026. Each source supports part of the overview; availability and prices can change.

• GitHub - bespokelabsai/curator: Synthetic data curation for post-training and structured data extraction · GitHub ↗github.com

Company website

github.com: product, pricing and policy pages

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

manual_review

Last checked 30 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.