Magika by Google

Magika identifies file content types using a machine-learning model. It examines a limited part of a file's contents and returns a predicted type, making it useful for developers building file-routing, ingestion or content-categorization pipelines. The original project URL now redirects to documentation hosted by Google Security Research, where the project describes a Rust command-line tool, a Python API and additional language bindings.

The documented classifier handles both binary and textual formats. Confidence thresholds and prediction modes help determine when to return a specific label and when to use a generic text or generic binary label. This can support choosing an appropriate downstream parser or scanner, but a predicted file type does not establish that a file is safe, authentic or free of malicious content.

A public browser demo was accessible and displayed a classification for its prefilled JavaScript example. The demo states that processing happens entirely in the browser. No custom text or file was submitted, and neither network behavior nor classification accuracy was independently audited. Published performance measurements therefore should not be treated as guarantees for an individual dataset.

The software is released under the Apache license. No paid hosted subscription or service-level commitment was established from the reviewed documentation; deployment and integration remain separate operational tasks. The known-limitations page says the release does not target polyglot detection, meaning files that can be interpreted as multiple formats need particular care. Despite the historical listing name, the project's disclaimer explicitly says it is not an official Google project and does not receive Google support. Evaluate it against representative files before relying on it in a production pipeline.