Documentation
How to call the Fronset API: quickstart, reference, errors, and what we do with your data.
The Fronset API is OpenAI-compatible: you point the SDK you already use at our
base URL and change the key. What differs is what model means — here it names a
task, a unit of work you define, and we choose the model that serves it. (The API
calls a task a capability in route and field names.)
Start with the Quickstart. The reference pages below are for when you need a specific field, header or error code.
The Terms of Service, Privacy Policy and Data Processing Agreement govern use of the API. Where anything on these pages differs from those, those govern.
- Quickstart
Point the OpenAI SDK you already use at Fronset and make a working call in under ten lines — including what to vault first and what `model` means here.
- Which Task Do I Use?
Find the catalogue task for the job you have (extract, score, summarise, classify, write, translate) and the benchmark page that shows how models do on it.
- Cookbook: Which Model Answered, and Why
Ten lines of Python: make one call on a task you create, then read which model answered it, what that cost, how much evidence we hold, and every model call behind it.
- Cookbook: Your Own Keys, How Comparison Runs, and How to Switch It Off
Two runs of the same 100 calls on your own provider key: one where Fronset compares models and one where you switch that off, with the model calls behind each.
- Cookbook: A Batch on a Settled Task
Summarise 50 documents in one batch on a catalogue task that has settled: upload, create, poll, download, and read which model answered each line, on which service tier, and what it cost.
- Cookbook: Define a Task of Your Own, Call It, and See Its Progress
Create a task of your own with its prompt template, output schema and model group, make two calls to it, and read where it stands: the headers on each call and the task's maturity.
- API Reference: Data Plane
The OpenAI-compatible chat completions and Batch endpoints, and the native endpoints: what matches OpenAI exactly, what differs, every llmbench extension field, every response header, and the errors you can get.
- API Reference: Control Plane
Everything you can do with a management-scope key: API keys, account, provider credentials, capabilities, routing policies, schemas, validators, webhooks, usage and billing.
- Prompt Variables and Capabilities
Define your own task: how prompt variables are substituted, how a capability's prompts, output schema and validators are created and edited, and what an edit changes.
- Lifecycle and What You Pay
How a task moves from exploring to a fixed price, who pays for the model calls we make along the way, and how to see what you were actually charged.
- Error Reference
Every error code the OpenAI-compatible plane can return, what it means, and what to do about it.
- Migrating from OpenAI
What changes and what does not when you repoint an OpenAI SDK at Fronset: base URL and key, the model field becoming a task slug, what fails loudly, what you gain, and the friction to expect.
- Data and Privacy
What we store, for how long, who else sees it, and what a deletion request does and does not reach — a plain-language index into the legal documents that govern.