What is datamatic?

datamatic is a command line tool for building multi step pipelines that call AI language models and pass structured data between the steps. Instead of writing custom code to call a model, check its output, and feed that output into another call, you describe the whole pipeline in a single configuration file, and datamatic runs it for you. Each step in a pipeline can call a different AI model, and datamatic works with several providers, including locally run models through Ollama or LM Studio, as well as cloud services like OpenAI, OpenRouter, and Google's Gemini. A step can ask a model to generate a fixed number of items, or it can repeat once for every item produced by an earlier step, referencing that item's fields directly inside its prompt. Steps can require the model to return data matching a defined structure, called a JSON schema, so the output is predictable and can be checked automatically rather than being free form text. Beyond calling models, datamatic can also run any command line tool as a step, load existing datasets from Hugging Face, and reshape or filter data between steps using an embedded version of the jq data processing language, so no separate jq installation is needed. Pipelines can also read values from environment variables, letting the same configuration file be reused with different models or settings. Typical uses described in the README include generating synthetic training data for fine-tuning other models, classifying documents through a structured multi step reasoning process, generating and reasoning about complex database queries, and combining image analysis with text generation. Output is written as structured JSON lines, with each line recording the prompt used, the model's response, and any values carried over from earlier steps for traceability. Installation is available through Homebrew, the Go install command, or by cloning the repository and building it from source. It requires Go to build from source, and an API key only if you choose a cloud based model provider.