Features

The whole architecture of Loom has a developer centric design which makes it easy to adapt it to your needs.

The Loom server provides a powerful REST API which can be used to manage all resources, including a Java client and a generated OpenAPI specification.

A built-in conversational assistant backed by a server-side agentic loop. It answers questions about your assets in natural language, with per-chat Skills, a persistent Memory bank, publishable Sessions and an optional isolated coding sandbox.

In addition to REST the server also provides a GraphQL API which allows you to fetch nested data in one step.

Loom comes with a powerful permission system. Permissions can be granted on an object level via users, groups and roles.

SSO IAM providers such as RedHat Keycloak, Auth0 and Okta are natively supported. No need for complicated plugins.

Existing assets can be referenced from disk. Alternatively files can be uploaded and stored locally (S3 backend planned).

A perceptual media fingerprint can be generated for video content. These fingerprints can be used for asset blacklisting and similarity search using nearest neighbor search.

Images can be resized and cropped. The focalpoint can be set on images in order to help generation of cropped images for responsive designs.

Contact-sheet thumbnails are automatically generated from video content by the Cortex thumbnail node.

Stored metadata can be indexed for full-text and vector search (Elasticsearch, Lucene and Qdrant integrations).

Loom comes with a system which allows you to tag assets, collections and content.

Beyond the structured models, Loom lets you attach custom metadata to any element it stores, together with user-generated content such as comments, reactions, ratings and annotations kept alongside your assets.

Selective data can be exported and re-imported in another instance.

Metadata from images, videos and documents is automatically extracted by Cortex nodes (hashes, transcripts, captions, OCR, faces, scenes …) and made accessible via the API.

Metadata is stored in PostgreSQL through a code-generated jOOQ layer, with Flyway-managed schema migrations.

Monitoring data is exposed via a Prometheus scrape endpoint, so you can track instance status from a Grafana dashboard.

Media processing runs on a fleet of Cortex worker daemons that register with Loom and execute node tasks it dispatches. Scale out by adding workers, and pin heavy node kinds (face detection, speech-to-text) to dedicated or GPU hardware.

The metaloom command line client talks to the same REST API as the web interface — script admin tasks, manage assets and run pipelines. Ships as a native binary (no JVM needed) or a runnable JAR.

Loom supports Kubernetes and ships a Helm Chart that helps you deploy it in your K8S cluster; Cortex workers deploy as scalable Deployments.

Extensions can be registered in order to extend the functionality of Loom, and custom Cortex processing nodes can be authored in Java or Python.

Assets can be referenced from an existing S3 bucket or you can choose to use a bucket to store newly uploaded assets.