@madewithremy/admin 0.1.59 → 0.1.60

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -601,9 +601,9 @@ export declare function rmWhereAll(ctx: AdminContext, params: RmWhereParams & {
601
601
  /**
602
602
  * Delete a whole data source — every version, every document, and every byte.
603
603
  *
604
- * Extraction caches survive (shared across sources by content hash), so
605
- * re-ingesting the same files into a new source costs no re-extraction.
606
- * Manage-plane only: running app code has no equivalent.
604
+ * Extracted text is shared across the app's sources by content hash, so a
605
+ * document's goes too unless another source in the app still holds the same
606
+ * bytes. Manage-plane only: running app code has no equivalent.
607
607
  *
608
608
  * @throws AdminApiError `data_source_not_found` (404).
609
609
  * @example
package/dist/prod.js CHANGED
@@ -7344,7 +7344,8 @@ Tuning a corpus:
7344
7344
 
7345
7345
  FREE \u2014 take effect on the next search, no rebuild:
7346
7346
  --rerank <true|false> Cross-encoder reranking (default true)
7347
- --rerank-model <id> Which cross-encoder reranks (live, no rebuild)
7347
+ --rerank-model <id> Which reranking model reranks (live, no rebuild);
7348
+ a chat model is refused, rerank_model_invalid
7348
7349
  --hybrid <true|false> Semantic + keyword matching (default true)
7349
7350
  --top-k <n> Default results per search
7350
7351
 
@@ -7379,7 +7380,16 @@ Tuning a corpus:
7379
7380
  one it lists; the error names the choices.
7380
7381
  Smaller vectors cut index memory, and dedicated
7381
7382
  capacity needs, roughly in proportion.
7382
- --extraction-model <id>
7383
+ --extraction-model <id> Reads PDFs and images; Office documents,
7384
+ spreadsheets, HTML and text never reach it.
7385
+
7386
+ Every model setting takes <model>@<routeKey> to hold that model's calls to
7387
+ one of its serving routes (\`routes\` in listModels), for a data-boundary
7388
+ need: \`--embedding-model text-embedding-3-large@openai\`. The pin is part of
7389
+ the setting, so it costs what the setting does, and it holds ingest and
7390
+ search alike. \`config\` shows it as stored; a route the model doesn't
7391
+ declare is refused with model_route_unavailable. The bare id routes
7392
+ automatically.
7383
7393
 
7384
7394
  Placement (shared pool vs dedicated capacity, see \`infra --help\`):
7385
7395
  --source defaults to "${DEFAULT_SOURCE}" and is created on first use, on the
@@ -7432,8 +7442,13 @@ Notes:
7432
7442
  recorded, e.g. '{"externalIdPrefix":"archive/2019/"}'). An empty filter
7433
7443
  is refused; to remove everything, delete the source.
7434
7444
  delete requires an explicit --source (no default) and refuses while documents
7435
- are still ingesting. Extraction caches survive deletion, so re-ingesting
7436
- the same files elsewhere costs no re-extraction.
7445
+ are still ingesting.
7446
+ Extracted text is kept by content hash and shared across the app's sources.
7447
+ It goes once no document in the app holds the same bytes (rm, rm --filter,
7448
+ delete), so a file still in another source keeps its extraction and
7449
+ re-adding it costs no re-extraction. Dropping a version (\`drop\`, or the
7450
+ sweep after \`promote\`) also deletes the extracted text of any extraction
7451
+ or image model no other version in the app uses.
7437
7452
  Re-vectorizing reuses stored extractions, so changing chunking never re-runs
7438
7453
  document extraction \u2014 only re-chunking and re-embedding.
7439
7454
 
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@madewithremy/admin",
3
- "version": "0.1.59",
3
+ "version": "0.1.60",
4
4
  "description": "Admin SDK + CLI for managing production Remy apps",
5
5
  "type": "module",
6
6
  "main": "./dist/index.js",