Augmentoolkit is an open-source toolkit on GitHub that turns your documents into domain-expert AI: it generates training datasets from raw text and uses them to fine-tune custom large language models that deeply know your material. The project has about 1,900 stars and is under active development, with updates continuing into 2026.
The workflow is deliberately simple: upload documents, press a button, and get back a fully trained custom LLM whose knowledge and behavior are shaped by your sources, whether that is the latest research in your field, internal documentation, or the lore of a fictional universe. Under the hood, Augmentoolkit creates high-quality instruction data from the texts, runs the fine-tuning, and can also produce a RAG-ready dataset plus an inference server if you prefer retrieval grounding. Importantly, it can operate largely offline on your own computer using local models for data generation on most hardware, with hosted APIs optional rather than required.
The toolkit grew out of its creator’s experiments in teaching open models specialized knowledge, and its community shares recipes for different model families and hardware setups. Support for modern local inference backends keeps experimentation accessible even without datacenter hardware.
Augmentoolkit is free and open source under the MIT license. Running it locally costs nothing beyond your own hardware; if you choose to use hosted model APIs for data generation or training, those providers charge their normal rates.
Augmentoolkit is for researchers, writers, indie hackers and small teams who want an LLM that genuinely knows their documents, and for anyone curious about fine-tuning who wants the dataset-creation heavy lifting automated.