@absolutejs/ai 0.6.5 → 0.6.7

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.
package/CHANGELOG.md CHANGED
@@ -6,6 +6,18 @@ This file is generated by `absolute-changelog` from the entries in
6
6
  `changelog/`. Edit an entry, not this file — and add new ones under
7
7
  `changelog/unreleased/`.
8
8
 
9
+ ## 0.6.7 — 2026-10-22
10
+
11
+ ### Fixed
12
+
13
+ - **Retain the requested model when validating legacy document checkpoints after provider recovery**
14
+
15
+ ## 0.6.6 — 2026-10-22
16
+
17
+ ### Changed
18
+
19
+ - **Recover approved providers in either direction across preparation and streaming; preserve checkpoints, normalize finite-key OpenAI schemas, and clear cancelled billing probes** (`AIInputError`, `PrepareAITextInputOptions`, `AIContextPolicy`, `streamWithProviderFallback`)
20
+
9
21
  ## 0.6.5 — 2026-10-22
10
22
 
11
23
  ### Changed
package/README.md CHANGED
@@ -593,3 +593,21 @@ polls or subscribes to actual progress, and binds actions to authenticated route
593
593
 
594
594
 
595
595
  `streamWithProviderFallback` accepts approved model IDs, an individually metered `stream(model)` callback, and an optional `onFallback` hook. It never restarts after a chunk or caller cancellation. `ProviderError.billingUnavailable` identifies provider billing failures; `notSent` identifies requests blocked locally. Billing suppression is scoped to the configured provider instance, with a five-minute probe interval.
596
+
597
+
598
+ ### Recovery across approved providers
599
+
600
+ Set `contextPolicy.providerFallback` on generation helpers and
601
+ `providerFallback` on `prepareAITextInput` / `prepareAITextInputStep` to cover
602
+ both token-capacity checks and generation. `models` accepts an array or a
603
+ function of the current model; the requested model runs first. Use a provider
604
+ router containing every approved model. Each alternate validates its own input
605
+ capacity. Optional `onFallback({ from, to, error })` reports the switch.
606
+
607
+ Recovery stops after any output chunk, including tool/usage chunks, and never
608
+ replays cancellation or accounting/application errors. Completed document
609
+ sections retain validated checkpoints across provider switches. Finite-key
610
+ OpenAI record schemas are translated to optional named properties without
611
+ removing value constraints. `isProviderError` works across package entrypoints;
612
+ `isPermanentAIProviderFailure` classifies terminal provider/preflight failures
613
+ for job runners. Callers own retry budgets, funding, and model approval.
package/changelog.json CHANGED
@@ -2,6 +2,32 @@
2
2
  "contract": 1,
3
3
  "name": "@absolutejs/ai",
4
4
  "releases": [
5
+ {
6
+ "changes": [
7
+ {
8
+ "kind": "fixed",
9
+ "summary": "Retain the requested model when validating legacy document checkpoints after provider recovery"
10
+ }
11
+ ],
12
+ "date": "2026-10-22",
13
+ "version": "0.6.7"
14
+ },
15
+ {
16
+ "changes": [
17
+ {
18
+ "kind": "changed",
19
+ "summary": "Recover approved providers in either direction across preparation and streaming; preserve checkpoints, normalize finite-key OpenAI schemas, and clear cancelled billing probes",
20
+ "symbols": [
21
+ "AIInputError",
22
+ "PrepareAITextInputOptions",
23
+ "AIContextPolicy",
24
+ "streamWithProviderFallback"
25
+ ]
26
+ }
27
+ ],
28
+ "date": "2026-10-22",
29
+ "version": "0.6.6"
30
+ },
5
31
  {
6
32
  "changes": [
7
33
  {