Immich uses a traditional client-server design, with a dedicated database for data persistence. The frontend clients communicate with backend services over HTTP using REST APIs. Below is a high level diagram of the architecture.
The diagram shows clients communicating with the server's API via REST. The server communicates with downstream systems (i.e. Redis, Postgres, Machine Learning, file system) through repository interfaces. Not shown in the diagram, is that the server is split into two separate containers `immich-server` and `immich-microservices`. The microservices container does not handle API requests or schedule cron jobs, but primarily handles incoming job requests from Redis.
All three clients use [OpenAPI](./open-api.md) to auto-generate rest clients for easy integration. For more information about this process, see [OpenAPI](./open-api.md).
:::
### Mobile App
The mobile app is written in [Flutter](https://flutter.dev/). It uses [Isar Database](https://isar.dev/) for a local database and [Riverpod](https://riverpod.dev/) for state management.
### Web Client
The web app is a [TypeScript](https://www.typescriptlang.org/) project that uses [SvelteKit](https://kit.svelte.dev) and [Tailwindcss](https://tailwindcss.com/).
The Immich CLI is an [npm](https://www.npmjs.com/) package that lets users control their Immich instance from the command line. It uses the API to perform various tasks, especially uploading assets. See the [CLI documentation](/docs/features/command-line-interface.md) for more information.
1. `redis`- Queue management for `immich-microservices`
### Immich Server
The Immich Server is a [TypeScript](https://www.typescriptlang.org/) project written for [Node.js](https://nodejs.org/). It uses the [Nest.js](https://nestjs.com) framework, with [TypeORM](https://typeorm.io/) for database management. The server codebase also loosely follows the [Hexagonal Architecture](<https://en.wikipedia.org/wiki/Hexagonal_architecture_(software)>). Specifically, we aim to separate technology specific implementations (`infra/`) from core business logic (`domain/`).
#### REST Endpoints
The server is a list of HTTP endpoints and associated handlers (controllers). Each controller usually implements the following CRUD operations:
- `POST` `/<type>` - **Create**
- `GET` `/<type>` - **Read** (all)
- `GET` `/<type>/:id` - **Read** (by id)
- `PUT` `/<type>/:id` - **Updated** (by id)
- `DELETE` `/<type>/:id` - **Delete** (by id)
#### DTOs
The server uses [Domain Transfer Objects](https://en.wikipedia.org/wiki/Data_transfer_object) as public interfaces for the inputs (query, params, and body) and outputs (response) for each endpoint. DTOs translate to [OpenAPI](./open-api.md) schemas and control the generated code used by each client.
### Microservices
The Immich Microservices image uses the same `Dockerfile` as the Immich Server, but with a different entrypoint. The Immich Microservices service mainly handles executing jobs, which include the following:
This list closely matches what is available on the [Administration > Jobs](/docs/administration/jobs-workers/#jobs) page, which provides some remote queue management capabilities.
All machine learning related operations have been externalized to this service, `immich-machine-learning`. Python is a natural choice for AI and machine learning. It also has some pretty specific hardware requirements. Running it as a separate container makes it possible to run the container on a separate machine, or easily disable it entirely.
Each request to the machine learning service contains the relevant metadata for the model task, model name, and so on. These settings are stored in Postgres along with other system configs. For each request, the microservices container fetches these settings in order to attach them to the request.
Internally, the machine learning service downloads, loads and configures the specified model for a given request before processing the text or image payload with it. Models that have been loaded are cached and reused across requests. A thread pool is used to process each request in a different thread so as not to block the async event loop.
All models are in ONNX format. This format has wide industry support, meaning that most other model formats can be exported to it and many hardware APIs support it. It's also quite fast.
Machine learning models are also quite _large_, requiring _quite a bit_ of memory. We are always looking for ways to improve and optimize this aspect of this container specifically.
See [Database Migrations](./database-migrations.md) for more information about how to modify the database to create an index, modify a table, add a new column, etc.
Immich uses [Redis](https://redis.com/) via [BullMQ](https://docs.bullmq.io/) to manage job queues. Some jobs trigger subsequent jobs. For example, Smart Search and Facial Recognition relies on thumbnail generation and automatically run after one is generated.
