Orinth
GitHubDownload

Overview

One workspace for computer vision, NLP, and LLM intelligence.

Orinth is a local studio where teams turn image and text datasets into usable model experiments — through labeling, preparation, training, testing, and inspection — and fine-tune, serve, and chat with language models in the same place.

Pronounced OR-inth (/ˈɔːrɪnθ/). Released under the Apache 2.0 license.

Workflow pillars

  • Organize — Create projects and datasets that keep image and text work scoped and understandable.
  • Prepare — Upload, label, annotate, split, preprocess, and version datasets without rewriting the originals. Dataset Studio supports image and text formats, imports from the Hugging Face Hub, and Data Recipes that build datasets from your own documents.
  • Train — Run task-compatible training jobs from prepared datasets using the local Model Library — including LoRA, QLoRA, and full LLM fine-tuning.
  • Test — Compare model behavior with task-aware metrics (accuracy, ROUGE, exact match, Dice, IoU) and per-item inspection.
  • Inspect — Keep model outputs, history, and dataset health visible for repeated iteration across every vision and NLP task.
  • Serve & chat — Export a fine-tuned model to GGUF, serve it locally, and chat with sampler controls, web search, and a reasoning view.

What's in this guide

  • Installation & Start — prerequisites, the desktop app, and running from source.
  • Workspace & Projects — how projects scope datasets, runs, and models.
  • Dataset Studio and Data Recipes — preparing and generating datasets.
  • Models & Training and LLM Fine-tuning & Chat — the Model Library, training runs, and language models.
  • Testing & Evaluation and Inference — measuring and running models.
  • NLP Capabilities — text classification, summarization, and question answering.
  • Model Catalog & Serving — the registry, custom uploads, GGUF export, and serving.
  • Guided Tours — the in-app walkthroughs.

Where your data lives

Everything Orinth generates — uploads, dataset versions, weights, logs, and prediction overlays — is written to local storage on the machine running it. There is no account, and no telemetry by default. The desktop app keeps its data under ~/Library/Application Support/; running from source keeps it under the directories named in your .env.

Orinth is a research and engineering instrument for measuring model behavior. It produces model outputs for human review — it does not provide clinical diagnosis or autonomous decision-making.