Ultrahost/ docs
All examples

Example 02

Experiment

Intelligent File Management

Store any kind of file, then make it findable, summarized, connected, and looked after.

Files are opaque until you give them context. This example keeps the bytes in one place and everything you know about them in another, then layers on search, similarity, summaries, and housekeeping so the pile stays useful.

Chapter one

When a file is uploaded

One upload, or many at once, sets off a chain of small jobs, each handled by a different block.

  1. 1
    File Server

    The file is stored

    The bytes go to the File engine. Everything else in the system points back to them.

  2. 2
    Document Database

    Its location and details are kept

    A JSON document records where the file lives, along with its name, type, size, and any other metadata.

  3. 3
    Authorization Layer

    Who can touch it is decided

    Rules declare who may read, edit, or delete the file, so access is settled from the first minute.

  4. 4
    LLM Plane

    It is summarized

    A language model reads the file and writes a short summary, which is saved with the rest of its metadata.

  5. 5

    At the same time

    It is made findable

    Three views of the file are prepared, so it can be reached in different ways.

    • Search Engine

      Indexed for contents

      Metadata and summary are indexed, so people can search for what is inside the file.

    • Vector DB

      Stored for similarity

      An embedding is saved, so files that resemble each other turn up together.

    • Graph Database

      Linked to its surroundings

      Edges connect the file to other files, documents, and related data.

Chapter two

When someone looks for a file

The same file can be reached from four directions.

Someone needs a file

A link, a few words, or an example they already have in hand.

  • By address

    Open it directly

    With its URL, the File engine serves the file safely.

    File Server

  • By tags or content

    Find what it says

    Search tags, metadata, and summaries to find files by what they contain.

    Search Engine

  • By similarity

    Find what looks alike

    Ask for files like this one, such as similar homes in a real estate catalog.

    Vector DB

  • By connections

    Find what it touches

    Start from one file and follow its links to related files and documents.

    Graph Database

Chapter three

While nobody is looking

A schedule keeps the pile in good shape without anyone asking.

A scheduled job wakes up

On a timer, with no one waiting for the result.

  • Clean up

    Purge what is forgotten

    Find old files that nothing references or uses anymore, and remove them.

    Cron

  • Back up

    Copy it somewhere safe

    Back files up to a third-party drive, so they survive a bad day.

    Cron

Make it yours

Variations worth trying

  • Skip summaries and similarity until search alone stops being enough.
  • Use the schedule only for backups if storage cost is not a concern.
  • Treat images, contracts, and recordings as separate flavors with their own metadata.

Starting point

Ask your coding agent for your version

Use this prompt as a starting point, then change whatever does not fit your idea.

Build a file manager where uploads are stored safely, described by metadata, searchable, summarized, and cleaned up automatically.

Use these Ultrahost building blocks, in this flow.

When a file is uploaded:
1. File Server: The bytes go to the File engine. Everything else in the system points back to them.
2. Document Database: A JSON document records where the file lives, along with its name, type, size, and any other metadata.
3. Authorization Layer: Rules declare who may read, edit, or delete the file, so access is settled from the first minute.
4. LLM Plane: A language model reads the file and writes a short summary, which is saved with the rest of its metadata.
5. It is made findable, at the same time:
   - Search Engine: Metadata and summary are indexed, so people can search for what is inside the file.
   - Vector DB: An embedding is saved, so files that resemble each other turn up together.
   - Graph Database: Edges connect the file to other files, documents, and related data.

When someone looks for a file:
- By address (File Server): With its URL, the File engine serves the file safely.
- By tags or content (Search Engine): Search tags, metadata, and summaries to find files by what they contain.
- By similarity (Vector DB): Ask for files like this one, such as similar homes in a real estate catalog.
- By connections (Graph Database): Start from one file and follow its links to related files and documents.

While nobody is looking:
- Clean up (Cron): Find old files that nothing references or uses anymore, and remove them.
- Back up (Cron): Back files up to a third-party drive, so they survive a bad day.

Treat this as a starting point. Keep the parts that fit my idea and adapt the rest.

Next example

Time-Series Analytics