Schema

Entity Diagram

How the codebase-knowledge entities connect. Auth, settings, ask sessions, benchmark/introspect runs, and the artifact tables sit in the same database but are intentionally omitted here so the graph reads cleanly. The full attribute list follows.

erDiagram
    REPO {
        string repo_uri
        string repo_head_sha
    }
    COMMIT {
        string git_sha PK
        string commit_message
        long   commit_timestamp
        long   commit_additions
        long   commit_deletions
    }
    AUTHOR {
        string author_name
        string author_email
    }
    DIR {
        string dir_path PK
    }
    FILE {
        string file_path PK
        keyword file_lang
        long    file_lines
        long    file_size
    }
    CODE {
        string code_name
        keyword code_kind
        long    code_line_start
        long    code_line_end
        keyword code_complexity
    }
    COMPONENT {
        string component_name
        keyword arch_layer
    }
    CHUNK {
        long   chunk_index
        string chunk_text
    }
    REPO ||--o{ COMMIT : "commits"
    REPO ||--|| DIR : "root"
    DIR  ||--o{ DIR  : "children"
    DIR  ||--o{ FILE : "contains"
    COMMIT }o--|| AUTHOR : "by"
    COMMIT ||--o{ COMMIT : "parents"
    COMMIT }o--o{ FILE   : "changed-files"
    FILE }o--o{ FILE : "imports"
    FILE ||--o{ CODE : "segments"
    CODE }o--o{ CODE : "calls"
    COMPONENT }o--o{ FILE      : "files"
    COMPONENT }o--o{ COMPONENT : "depends-on"
    CHUNK }o--|| FILE      : "of file"
    CHUNK }o--|| COMPONENT : "of component"

Two Databases per Instance

Every Noumenon instance hosts two kinds of database. Per-repo databases hold facts derived from a specific repository: commits, files, code segments, components, the import graph, the TF-IDF index. One per imported repo. Identity is derived from the repo path or URL. Schema files: core.edn, architecture.edn, synthesis.edn, provenance.edn.

The noumenon-internal meta database holds cross-cutting facts about Noumenon itself: introspect runs, ask sessions and feedback, named-query artifacts, prompts, auth tokens, settings. One per instance, fixed name. A prompt change affects every repo, so its history belongs in one place rather than fragmented per-repo. Schema files: ask.edn, benchmark.edn, introspect.edn, artifacts.edn, auth.edn, settings.edn.

Both run on the same Datomic Local storage and share the same schema-attribute namespace. The split is conceptual: facts about your code versus facts about how Noumenon answered questions about your code.

Traversing in Both Directions

Datomic refs are stored once but queried both ways. Forward, you name the attribute. Backward, you prefix the name with an underscore.

;; Forward: which files does this file import?
[?file :file/imports ?dep]

;; Backward: which files import this file?
[?file :file/_imports ?dependent]

;; Forward: which segments does this code segment call?
[?caller :code/calls ?callee]

;; Backward: which segments call this one?
[?callee :code/_calls ?caller]

Every :db.type/ref in the schema below is a candidate for backward traversal. The Ask agent and the named-query catalog use both directions extensively.

Attribute Reference

249 attributes across 28 namespaces, parsed from resources/schema/ in the noumenon repo. Filter by name or by description text.