Documentation

Everything you need to integrate ReOpenly, an OpenAI-compatible API for datasets, fine-tuning and inference.

01QuickstartKey, dataset, fine-tune, served model. Five requests, start to finish.02AuthenticationCreate a key, send it as a bearer token, revoke it the moment you suspect it.03Inference APIModels, chat, completions, streaming and context limits, through the unmodified OpenAI SDK.04Training jobsStart, watch and stop fine-tunes, with the real state and attempt tables.05Hosted modelsPublish a finished fine-tune under a stable name, and share it or make it public.06Errors and retriesStable codes and Retry-After, and which failures are never worth retrying.07DatasetsChat CRUD, provenance, filtering, pagination, and durable CSV imports.08Pricing and creditsLive estimates, authorization ceilings, statements, and credit checkout.09Endpoint indexCustomer REST resources and the full API reference, with every parameter and response.10Quickstart for embeddingsTraining pairs, a tuned embedding model, hosted and called through the OpenAI embeddings API.11Playground and SearchCompare texts and rank a whole dataset with a hosted embedding model before you build on it.

One base URL

Everything is scoped to a project, and the owner and project are part of the URL. That is what lets a client keep a stable model name while the platform resolves the project, the payer and the model behind it.

https://reopenly.com/api/projects/{owner_slug}/{project_slug}/openai/v1

The global playground uses the same project-scoped API: once a project is selected its page URL is /app/playground?project=owner/project, so a refresh or a shared link keeps the scope. There is no unscoped inference route.