Self-hosted, Self-Improving, Federated Everything App — dev management, agent orchestration, persistent mind, digital twin, second brain, and creative pipeline from a single dashboard over Tailscale to port the user into the OS
JavaScript ai-agentsdashboarddeveloper-toolsdigital-twinquantified-selfsecond-brainself-hostedtailscale
/**
* Media asset index — pure row transforms.
*
* The index (#1000) is a DERIVED, queryable mirror of media that physically
* lives on disk: generated images (data/images/*.png + .metadata.json sidecars)
* and generated videos (data/videos/*.mp4, tracked in data/video-history.json).
* Those files stay authoritative; this module just turns a disk record into the
* `media_assets` row shape so the reconcile pass + the live completed-hook can
* never drift on what a row looks like. No I/O here.
*
* A row is `{ mediaKey, kind, ref, data, createdAt }`:
* - kind/ref → the shared `<kind>:<ref>` vocabulary (mediaItemKey.js)
* - mediaKey → `<kind>:<ref>` (the PK)
* - data → the full metadata record, stored verbatim in JSONB
* - createdAt → bind-safe TIMESTAMPTZ for the queryable column
*/
import { itemKey } from '../../lib/mediaItemKey.js';
import { mirrorTimestamp } from '../../lib/pgTimestamp.js';
// The one place each kind's ref is turned into a media_key. Both the row
// builders (which WRITE rows) and the delete hooks (which REMOVE them) derive
// their key here, so an unindex can never miss the row its upsert wrote.
// Returns null for an unusable ref, so callers can filter/no-op on it.
const mediaKeyFor = (kind, ref) => (typeof ref === 'string' && ref ? itemKey({ kind, ref }) : null);
/** media_key for a generated image. Its ref is the gallery FILENAME. */
export const imageMediaKey = (filename) => mediaKeyFor('image', filename);
/** media_key for a generated video. Its ref is the job ID, not the filename. */
export const videoMediaKey = (id) => mediaKeyFor('video', id);
/**
* Build an index row for a generated image. `item` is a gallery entry as
* produced by imageGen listGallery() — `{ filename, createdAt, ...sidecar }` —
* or any object carrying at least a `filename`. Returns null when there's no
* usable ref (so callers can filter).
*/
export function imageToRow(item, { now } = {}) {
const ref = item?.filename;
const mediaKey = imageMediaKey(ref);
if (!mediaKey) return null;
const fallback = now || new Date().toISOString();
return {
mediaKey,
kind: 'image',
ref,
data: item,
createdAt: mirrorTimestamp(item.createdAt, fallback),
};
}
/**
* Build an index row for a generated video. `entry` is a video-history record —
* `{ id, filename, createdAt, ... }`. The video's ref in the `<kind>:<ref>`
* vocabulary is its job id (matches how mediaCollections stores video items),
* NOT the filename. Returns null when there's no usable id.
*/
export function videoToRow(entry, { now } = {}) {
const ref = entry?.id;
const mediaKey = videoMediaKey(ref);
if (!mediaKey) return null;
const fallback = now || new Date().toISOString();
return {
mediaKey,
kind: 'video',
ref,
data: entry,
createdAt: mirrorTimestamp(entry.createdAt, fallback),
};
}
// Card fields a compact list row keeps: identity, file addressing, lineage
// badges and the chips MediaCard renders. Everything else — full prompts,
// negative prompts, provider payloads — stays on the record and is read on
// demand (`/gallery/lookup`, `/video-gen/history/:id`).
const COMPACT_FIELDS = ['id', 'filename', 'path', 'thumbnail', 'createdAt', 'hidden', 'width', 'height',
'appId', 'model', 'modelId', 'mode', 'seed', 'steps', 'numFrames', 'fps', 'fileSizeBytes', 'renderMs', 'loraFilenames',
'lora_filenames', 'loraPaths', 'lora_paths', 'stitchedFrom', 'upscaledFrom', 'extractedFromVideoId',
'cleanedFrom', 'autoCleaned', 'regenerated', 'watermarkRemoved'];
export const COMPACT_PROMPT_CHARS = 240;
// Scalars the list predicates read (db.js assetFilter) that are not card
// fields. A compact source keeps them so scope filters match exactly as they
// do over the full record; compactSourceToRow drops them.
const COMPACT_FILTER_FIELDS = ['universeId', 'entryCategory', 'entryKind'];
export const COMPACT_SOURCE_FIELDS = [...COMPACT_FIELDS, ...COMPACT_FILTER_FIELDS];
// Internal key carrying the chosen prompt, bounded to one character past the
// preview so the preview (and its ellipsis) is identical to one cut from the
// full prompt. Never returned to a client.
export const COMPACT_PREVIEW_KEY = '_promptPreviewSource';
const nonBlank = value => typeof value === 'string' && value.trim() ? value : null;
/**
* Bounded projection of a full record: card fields, filter-only scalars and
* the bounded preview source. This is what a compact no-search list sends to
* SQL in place of the full video-history record (#9676); db.js builds the
* same shape for indexed images in SQL.
*/
export function compactSource(data) {
if (!data || typeof data !== 'object') return data;
const source = {};
for (const field of COMPACT_SOURCE_FIELDS) if (data[field] !== undefined) source[field] = data[field];
// Re-projecting a cached source must preserve whitespace-only prefixes: the
// original nonblank text may lie beyond the preview boundary.
const prompt = nonBlank(data.prompt) || nonBlank(data.metadata?.prompt)
|| (typeof data[COMPACT_PREVIEW_KEY] === 'string' ? data[COMPACT_PREVIEW_KEY] : null);
if (prompt) source[COMPACT_PREVIEW_KEY] = prompt.slice(0, COMPACT_PROMPT_CHARS + 1);
return source;
}
/** The client-facing compact row for a compact source; internal fields are dropped. */
export function compactSourceToRow(source) {
if (!source || typeof source !== 'object') return source;
const row = { compact: true };
for (const field of COMPACT_FIELDS) if (source[field] !== undefined) row[field] = source[field];
const prompt = source[COMPACT_PREVIEW_KEY];
if (typeof prompt === 'string' && prompt) {
row.prompt = prompt.length > COMPACT_PROMPT_CHARS ? `${prompt.slice(0, COMPACT_PROMPT_CHARS).trimEnd()}…` : prompt;
}
return row;
}
/**
* Opt-in list projection (#8292). `compact: true` marks the row as a PREVIEW:
* its prompt is cut to a display label, so a consumer must hydrate the full
* record before editing it or handing its prompt/settings to a generation
* action. Search and counts still run over the full stored metadata.
*/
export const compactGalleryRecord = data => compactSourceToRow(compactSource(data));