panorama/docs/src/index.md

- CLI API

  - `panorama install` : Install a script into the database
  - `panorama show` : Show the information about a particular node, by ID

- Database of nodes + code

  - Code

    - Code is all interface-based. Interface implementations give specific ways
      behaviors are implemented

    - Standard interfaces:

      - Indexed

  - Node definition

    ```typescript
    type Node = {
      metadata: KVStore,
      interfaces: Set<InterfaceURI>,
      edges: Set<(EdgeType, Edge)>,
    }
    ```

    - Within the metadata kv store, keys are namespaced, according to the
      application name. Application names must be registered ahead of time

    - Nodes implement interfaces, the code of which is fetched from the internet
      and validated

    - Nodes are namespaced, although namespacing isn't the primary hierarchy
      protocol.

      - This is needed because the database itself has a privileged namespace
        - The privileged namespace is for keeping track of database state, for
          example which scripts are trusted

    - Hooks:

      - Hooks consist of a (pattern, type, callback). The pattern is a metadata
        matcher, that determines if the node should be executed upon. The
        callback is Javascript code executed in the app's execution context.

      - Type determines what kind of callback to call.

        - Ensures callbacks: these make sure that a node has

      - Need to somehow figure out how to represent partially executed
        processes? i.e if a video indexer needs to take time to go through the
        video, the nodes should be available and committed, but somehow
        indicate that one of its processes isn't done
        - Maybe this should be a flag on the `NodeMeta` table

  - Database service:

    - IFTTT Events

      - Local vs. distributed IFTTT events?
        - Local means that ur particular device is repsonsible for running it
        - Distributed means any device can consume the event???
          - is this even a good idea? Maybe just have one dedicated worker
            device. For example, may not want to run certain automation tasks
            on phone, while others may only be able to be run on phone

    - Cron tasks

      - Sync protocol:
        - After a modification, the modification gets written to a queue
        - CRDT?

- Other privileged concepts:

  - User: a pub/priv key pair

- Sample apps

  - Todos app

    - Metadata keys (these are somewhat based on Taskwarrior)

      - `status`: Enum{Pending | Deleted | Completed | Waiting | Recurring}
        - Task status
      - `description`: String
        - Description
      - `start`: DateTime
      - `due`: DateTime{.due >= .start}
      - `depends`: NodeRef

    - Interfaces:

      - `addTodo()`

  - Content / blob storage

    - Interfaces:

      - `Content`
        - `upload() -> Writer`
        - `get(node_id : NodeId) -> Reader`
      - `Writer`
        - `next_chunk(Vec<u8>)`
      - `Reader`
        - `next_chunk() -> Vec<u8>`

  - Personal journaling (like Logseq)

  - Weight / fitness tracking

  - Calendar app (store events)

  - Git app (store code)

  - File backup

  - Chat / Mail?

  - Photo viewer + indexing using ML models for face detection / keywords

  - Not sure yet:

    - Automatic sharing of database nodes? This seems like a bad idea for
      multiple reasons. Maybe better to just have a specific app for
      doing permission models

- Other concerns

  - Local protection using libsecret?

  - Migrations?

    - If an application needs to migrate its data format, should we formalize
      this?
      - Data migrations that don't need to create / delete nodes, only metadata

  - Have the frontend also shipped with the backend?

    - Subset of React components to typecheck against

  - Aggregated queries and group by? Joins?