What is ReOpenly?
ReOpenly is a managed platform for adapting open models to a business’s data, including embedding models for search and retrieval-augmented generation (RAG). Teams can turn search feedback or prepared question-and-passage examples into training data, train and host a model, then call it through an OpenAI-compatible API. It is for teams whose applications search their own documents or databases and want to tailor retrieval without managing the training and hosting infrastructure themselves.
ReOpenly Features:
Search Feedback: Send queries and the results your application showed to ReOpenly, then review them and mark which passages answer each query.
Dataset Import: Upload CSV or JSONL examples with a question, an answering passage and an optional related passage that does not answer it.
Training Controls: Choose datasets or tags for a run and adjust settings such as learning rate, vector size, input length and query instructions.
Model Evaluation: Use the Playground to inspect similarity, explore vector groupings, test search and compare tuned models with a base model.
Hosted Embeddings: Host a trained model under a chosen name and call its embeddings endpoint using the OpenAI SDK, cURL or JavaScript.
Run Cost Preview: Review a training run’s data split and maximum estimated cost before starting; training is billed for measured GPU time.
ReOpenly Benefits:
More Relevant Retrieval: Training with examples from real searches gives teams a way to align results with what answers questions in their own collection.
Safer Model Changes: Testing against searches held aside from training helps teams judge whether a tuned model is useful before switching over.
Less Infrastructure Work: Managed training and hosting let teams adapt an embedding model without setting up their own GPU training and serving systems.
Clearer Cost Decisions: Seeing the run’s maximum estimated cost in advance helps teams decide whether to proceed, while an unstarted training run is free.
More Control Over Models: Downloading trained adapters gives teams a copy of their model artifacts rather than making access depend entirely on the hosted endpoint.
Use Cases:
Knowledge Base RAG: Tune retrieval for an assistant that searches help-center articles so it can surface passages that directly address customer questions.
Internal Document Search: Use reviewed employee searches to train a retriever for company documents where generally similar passages are not always useful answers.
Support Search Tuning: Collect the queries and displayed results from a support portal, then turn reviewed helpful results into training examples.
Database Retrieval: Adapt search for a large database when an application’s current embedding model returns related records that do not answer the query.
RAG Model Evaluation: Compare a tuned embedding model with its base model on held-aside searches before deciding whether to update a retrieval pipeline.

