@chipmobilesdk/rn-ai 0.2.0 → 0.3.0

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
@@ -1,55 +1,158 @@
1
- # Changelog — @chipmobilesdk/rn-ai
2
-
3
- ## [0.2.0] — 2026-11-01
4
-
5
- Fixes a chat or quota call that could hang forever, and errors that never
6
- reached the app. **No code change is required.**
7
-
8
- ### Fixed
9
-
10
- - **Every network call has a time limit.** The chat stream had none, and
11
- neither did the config or quota fetch, so on a stalled mobile connection
12
- `useAiChat` stayed in `loading`/`streaming` and `useAiQuota` in `loading`
13
- until the user gave up — and with no cached config, one hung fetch blocked
14
- every call. New `timeouts: { connectMs, idleMs, configMs, quotaMs }` option
15
- (defaults 30 s, 30 s, 10 s, 10 s). The idle limit restarts on every chunk.
16
- - **No error is swallowed.** Every chat path ended in `.catch(() => undefined)`,
17
- so a transport that threw (an invalid URL from remote config, for example)
18
- left the UI in `loading` with nothing to show. Such failures now arrive as
19
- `AiError('internal')`. The remote `recovery` block is type-checked when
20
- parsed, so a mistyped value can no longer throw inside the XHR callback.
21
- - **A cancelled or replaced request stays silent.** A token or config failure
22
- that landed after `cancel()` (or after the next `send()`) still called
23
- `onError` and flipped that screen to `error`. The client now routes every
24
- callback through one guard (nothing after cancel, exactly one of
25
- `onDone`/`onError`), and `useAiChat` and `useAiQuota` drop callbacks from
26
- older requests.
27
-
28
- ### Changed
29
-
30
- - A stalled stream, a timed-out quota fetch and a dropped connection (status 0)
31
- report `upstream-error` with `recoverable: true`; status 0 was
32
- `recoverable: false`.
33
-
34
- ### Added
35
-
36
- - `AiClientConfig.timeouts` and the `AiTimeouts` type.
37
-
38
- ## 0.1.0 — 2026-07-05
39
-
40
- Initial release (feature `005-azure-ai-gateway`).
41
-
42
- - `createAiClient` — config-driven gateway client: cache-first app behavior
43
- config (ETag/TTL), token acquisition via `@chipmobilesdk/rn-auth`
44
- (`resourceKey`), SSE streaming over zero-dependency XHR transport.
45
- - Multimodal chat requests: `text`, `image` (base64), `audio` (wav/mp3).
46
- - Silent endpoint re-route on designated `404`/`410` signals (max 1 retry);
47
- one automatic fresh-token retry on `401`; no automatic retry on `429`.
48
- - `useAiChat` (idle/loading/streaming/done/error/rate-limited + accumulated
49
- text + `retryAt`) and `useAiQuota` hooks.
50
- - Typed `AiError` codes and `ai.*` message keys with English defaults.
51
- - Per-app kill switch honored (`features.aiEnabled`).
52
-
53
- Adoption: additive — no changes required in existing apps. Requires an
54
- `ai-gateway` resource entry in the rn-auth config and a registered app on the
55
- gateway platform.
1
+ # @chipmobilesdk/rn-ai
2
+
3
+ ## 0.3.0
4
+
5
+ ### Minor Changes
6
+
7
+ - One diagnostic stream: install a sink with `installSdkSink` from
8
+ `@chipmobilesdk/rn-core` and it receives what every package observes (events,
9
+ failures, warnings) as `SdkDiagnostic`s, redacted. Listeners are notified through
10
+ rn-core, so a listener that throws no longer stops the others nor fails the call that
11
+ notified, and its error is reported rather than swallowed. ([#17](https://github.com/operationd8979/chip-mobile-sdk/pull/17))
12
+
13
+ - rn-ads: a banner's auto-refresh is a new cycle, so its `loaded` and `impression`
14
+ events are no longer deduplicated away; the dedupe memory is bounded. A throwing
15
+ status, consent or placement listener is reported as an `internal_error` event.
16
+ - rn-i18n: new `onDiagnostic` config receives every diagnostic (`I18nDiagnostic`,
17
+ with a stable `code`); without it, warnings go to `console.warn` in development
18
+ builds only. The warn-once memory is bounded.
19
+ - rn-chart: now depends on rn-core for its diagnostics.
20
+ - rn-payment: status, entitlement and pending listeners report a throw to the logger
21
+ instead of swallowing it.
22
+
23
+ - The package's errors extend `SdkError` from the new `@chipmobilesdk/rn-core`
24
+ dependency: `isSdkError` recognises them alongside every other package's, and each
25
+ carries `pkg`, `code` and `retryable`. Names and codes are unchanged. ([#17](https://github.com/operationd8979/chip-mobile-sdk/pull/17))
26
+
27
+ - rn-ai and rn-camera: `retryable` is the new name of `recoverable`, which stays as a
28
+ deprecated alias.
29
+ - rn-auth: `retryable` is true for `network-unavailable`. `redact` also removes URLs,
30
+ e-mail addresses and UUIDs.
31
+ - rn-ads: `redact` also removes JWTs and e-mail addresses, and reduces an `AdsError`
32
+ to its code.
33
+ - rn-ai: an error message quoted from the gateway is pattern-redacted.
34
+ - rn-camera: `redact` also removes JWTs, e-mail addresses and UUIDs.
35
+ - rn-theme: `ThemeConfigError` has `code: 'invalid_config'` and `ThemeContrastError`
36
+ `code: 'contrast'` (new `ThemeErrorCode` type).
37
+ - rn-i18n: `I18nConfigError` has `code: 'invalid_config'` and `UnitConversionError`
38
+ `code: 'unit_conversion'` (new `I18nErrorCode` type).
39
+
40
+ - Timers, retries and bounds now come from `@chipmobilesdk/rn-core`, and the drift
41
+ between the copies is fixed. ([#17](https://github.com/operationd8979/chip-mobile-sdk/pull/17))
42
+
43
+ - rn-ads: a retry policy out of bounds throws `retry_out_of_bounds`, as documented,
44
+ instead of `timeout_out_of_bounds`; a placement's `minRequestIntervalMs` out of
45
+ bounds throws `invalid_config`.
46
+ - rn-payment: backoff validates `initialDelayMs` (100 ms to 60 s, not above
47
+ `maxDelayMs`) and `factor` (1 to 10); `factor: 0` used to retry at once, forever.
48
+ Every timeout, backoff or interval out of bounds now throws the new
49
+ `PAYMENT_INVALID_CONFIG` instead of `PAYMENT_INVALID_CATALOG`, which is about
50
+ products.
51
+ - rn-ai: concurrent calls share one behavior-config request, so a cold start makes
52
+ one instead of one per call. A silent retry after a config reload waits
53
+ `recovery.retryBackoffMs`, which was parsed but never applied.
54
+
55
+ - Hooks read through `useSyncExternalStore` (rn-core's `useExternalStore`) and
56
+ re-render only when what they return changes. ([#17](https://github.com/operationd8979/chip-mobile-sdk/pull/17))
57
+
58
+ - rn-auth: `useProfile` and `useAccessTier` no longer re-render for a status change,
59
+ and the provider's context no longer changes on every transition. New
60
+ `useAuthSelector`. `auth.getState()` returns the same frozen object until the state
61
+ changes, instead of a copy per call.
62
+ - rn-payment: `useEntitlement` re-renders only when its own entitlement changes; the
63
+ engine keeps unchanged entitlement objects across syncs and does not notify for a
64
+ sync that changed nothing.
65
+ - rn-ads: `usePlacementState` and `useAdsStatus` read tear-free; `useAdsEvents` keeps
66
+ one subscription when given an inline listener.
67
+ - rn-i18n: `useTranslation` returns the same `t` until the locale changes, and
68
+ `useLocale`'s `setLocale` is stable.
69
+ - rn-ai: new `useAiBehaviorState`, with `client.subscribe` and a stable
70
+ `client.getBehaviorState()`.
71
+
72
+ - Every package's own copy is an rn-core message catalog that an app can translate. ([#17](https://github.com/operationd8979/chip-mobile-sdk/pull/17))
73
+
74
+ - rn-i18n: new `runtimeEntriesFromCatalog(catalog)` registers any SDK catalog in one
75
+ line, converting `{name}` variables to `{{name}}`.
76
+ - rn-ai: new `messageKeyForError(error)` gives the copy for every error code, and
77
+ `aiMessageSet`. The input placeholder default is neutral ("Type a message…"), and
78
+ two keys were added for errors that had none (`ai.error.signIn`, `ai.error.generic`).
79
+ - rn-auth: new `authMessages`, the catalog split into user and developer copy.
80
+ - rn-payment: new `paymentMessages`.
81
+ - rn-chart: summaries, announcements, adjustable-action labels, built-in state text
82
+ and legend hints come from `chartStrings`; pass `strings` to translate them. The
83
+ `accessibility.summaryFormatter` prop, declared before, now replaces the summary.
84
+
85
+ ### Patch Changes
86
+
87
+ - Updated dependencies ([#17](https://github.com/operationd8979/chip-mobile-sdk/pull/17), [#17](https://github.com/operationd8979/chip-mobile-sdk/pull/17), [#17](https://github.com/operationd8979/chip-mobile-sdk/pull/17)):
88
+ - @chipmobilesdk/rn-core@0.1.0
89
+
90
+ ## 0.2.1
91
+
92
+ ### Patch Changes
93
+
94
+ - Packaging only; no code change. The package no longer declares `engines.node`: a
95
+ React Native package does not run on Node, so the field only produced install
96
+ warnings in apps on another Node version. rn-i18n's command-line tooling does run
97
+ on Node; its README's compatibility table states the version it needs. ([#14](https://github.com/operationd8979/chip-mobile-sdk/pull/14))
98
+
99
+ ## [0.2.0] — 2026-11-01
100
+
101
+ Fixes a chat or quota call that could hang forever, and errors that never
102
+ reached the app. **No code change is required.**
103
+
104
+ ### Fixed
105
+
106
+ - **Every network call has a time limit.** The chat stream had none, and
107
+ neither did the config or quota fetch, so on a stalled mobile connection
108
+ `useAiChat` stayed in `loading`/`streaming` and `useAiQuota` in `loading`
109
+ until the user gave up — and with no cached config, one hung fetch blocked
110
+ every call. New `timeouts: { connectMs, idleMs, configMs, quotaMs }` option
111
+ (defaults 30 s, 30 s, 10 s, 10 s). The idle limit restarts on every chunk.
112
+ - **No error is swallowed.** Every chat path ended in `.catch(() => undefined)`,
113
+ so a transport that threw (an invalid URL from remote config, for example)
114
+ left the UI in `loading` with nothing to show. Such failures now arrive as
115
+ `AiError('internal')`. The remote `recovery` block is type-checked when
116
+ parsed, so a mistyped value can no longer throw inside the XHR callback.
117
+ - **A cancelled or replaced request stays silent.** A token or config failure
118
+ that landed after `cancel()` (or after the next `send()`) still called
119
+ `onError` and flipped that screen to `error`. The client now routes every
120
+ callback through one guard (nothing after cancel, exactly one of
121
+ `onDone`/`onError`), and `useAiChat` and `useAiQuota` drop callbacks from
122
+ older requests.
123
+
124
+ ### Changed
125
+
126
+ - A stalled stream, a timed-out quota fetch and a dropped connection (status 0)
127
+ report `upstream-error` with `recoverable: true`; status 0 was
128
+ `recoverable: false`.
129
+
130
+ ### Added
131
+
132
+ - `AiClientConfig.timeouts` and the `AiTimeouts` type.
133
+
134
+ ## 0.1.1 — 2026-08-01
135
+
136
+ ### Changed
137
+
138
+ - The package now ships its agent skill (`skills/`), which consuming apps
139
+ copy in with `scripts/sync-sdk-skills.mjs`. No code change.
140
+
141
+ ## 0.1.0 — 2026-07-05
142
+
143
+ Initial release (feature `005-azure-ai-gateway`).
144
+
145
+ - `createAiClient` — config-driven gateway client: cache-first app behavior
146
+ config (ETag/TTL), token acquisition via `@chipmobilesdk/rn-auth`
147
+ (`resourceKey`), SSE streaming over zero-dependency XHR transport.
148
+ - Multimodal chat requests: `text`, `image` (base64), `audio` (wav/mp3).
149
+ - Silent endpoint re-route on designated `404`/`410` signals (max 1 retry);
150
+ one automatic fresh-token retry on `401`; no automatic retry on `429`.
151
+ - `useAiChat` (idle/loading/streaming/done/error/rate-limited + accumulated
152
+ text + `retryAt`) and `useAiQuota` hooks.
153
+ - Typed `AiError` codes and `ai.*` message keys with English defaults.
154
+ - Per-app kill switch honored (`features.aiEnabled`).
155
+
156
+ Adoption: additive — no changes required in existing apps. Requires an
157
+ `ai-gateway` resource entry in the rn-auth config and a registered app on the
158
+ gateway platform.
package/README.md CHANGED
@@ -1,151 +1,159 @@
1
- # @chipmobilesdk/rn-ai
2
-
3
- AI gateway client for ChipMobileSdk-family React Native apps. Streams
4
- multimodal (text + image + audio in, text out) AI responses through the
5
- platform gateway with per-user rate limiting — apps never talk to Azure
6
- OpenAI directly and hold no AI credentials.
7
-
8
- > **License:** `UNLICENSED`. Published publicly for use by the owner's
9
- > applications; no open-source license is granted.
10
-
11
- - **Version**: `0.2.x` (see [CHANGELOG](CHANGELOG.md))
12
- - **Entrypoint**: `@chipmobilesdk/rn-ai`
13
- - **Native code**: none
14
-
15
- - Contract: [rn-ai-public-api](../../specs/005-azure-ai-gateway/contracts/rn-ai-public-api.md)
16
- - Gateway API: [gateway-api](../../specs/005-azure-ai-gateway/contracts/gateway-api.md)
17
- - Behavior config: [app-behavior-config](../../specs/005-azure-ai-gateway/contracts/app-behavior-config.md)
18
-
19
- ## Install
20
-
21
- ```bash
22
- npm install @chipmobilesdk/rn-ai @chipmobilesdk/rn-auth
23
- ```
24
-
25
- Peers: `react >=19`, `react-native >=0.85`, `@chipmobilesdk/rn-auth >=0.2.0`.
26
- **Zero runtime dependencies**; streaming uses React Native's standard
27
- `XMLHttpRequest` (no native code, Hermes-compatible, iOS + Android).
28
-
29
- ## Setup
30
-
31
- 1. Add the gateway resource to your rn-auth config:
32
-
33
- ```ts
34
- resources: [
35
- { key: 'ai-gateway', scopes: ['api://aiapp-gateway/AI.Access'] },
36
- ]
37
- ```
38
-
39
- 2. Create the client (one per app):
40
-
41
- ```ts
42
- import { createAiClient } from '@chipmobilesdk/rn-ai';
43
- import { auth } from './auth/setup';
44
-
45
- export const aiClient = createAiClient({
46
- appId: 'scan-food',
47
- configUrl: 'https://<config-host>/appconfig/config/scan-food.json',
48
- resourceKey: 'ai-gateway',
49
- auth,
50
- // storage: inject a persistent AiConfigCache for offline launches
51
- });
52
- ```
53
-
54
- 3. Use the hooks:
55
-
56
- ```tsx
57
- import { useAiChat, useAiQuota } from '@chipmobilesdk/rn-ai';
58
- import { aiClient } from './ai/setup';
59
-
60
- const { send, cancel, status, text, error, retryAt } = useAiChat(aiClient);
61
- const { quota } = useAiQuota(aiClient);
62
-
63
- send({
64
- messages: [{
65
- role: 'user',
66
- content: [
67
- { type: 'text', text: 'What dish is this? Estimate calories.' },
68
- { type: 'image', mimeType: 'image/jpeg', dataBase64: photoBase64 },
69
- ],
70
- }],
71
- });
72
- ```
73
-
74
- ## Runtime behavior
75
-
76
- - **Config-driven endpoint**: resolved lazily on the first quota or chat
77
- request. A fresh cache entry is used immediately; a stale entry is used
78
- immediately and revalidated in the background with its ETag. With no cache,
79
- the first operation waits for the config request, for at most `configMs`.
80
- Designated `404`/`410`
81
- responses invalidate config and retry against the refreshed endpoint.
82
- - **Auth**: acquires tokens via rn-auth (`requestAccessToken('ai-gateway')`);
83
- on `401`, reacquires a token and retries once.
84
- - **Timeouts**: every network call is bounded. Override any of them with
85
- `createAiClient({ timeouts: { ... } })`:
86
-
87
- | | Default | Fails with |
88
- |---|---|---|
89
- | `connectMs` — chat request sent → headers | 30 s | `upstream-error`, `recoverable: true` |
90
- | `idleMs` — silence between two stream chunks | 30 s | `upstream-error`, `recoverable: true` |
91
- | `configMs` — behavior-config fetch | 10 s | stale cache if any, else `config-unavailable` |
92
- | `quotaMs` — quota fetch | 10 s | `upstream-error`, `recoverable: true` |
93
-
94
- The gateway sends headers only after the model answers, so `connectMs` also
95
- covers time to first token. A dropped connection (HTTP status 0) is likewise
96
- `upstream-error` with `recoverable: true`.
97
- - **Callbacks**: each `chat()` delivers exactly one of `onDone` / `onError`,
98
- and nothing after `cancel()`. An unexpected exception anywhere on the way
99
- (including in your own `onDelta`) is reported as `internal`, never dropped.
100
- - **Rate limits**: `429` surfaces as `status: 'rate-limited'` with `retryAt`;
101
- the SDK never hammers and never enforces limits locally (display only).
102
- - **Typed errors** (`AiError.code`): `auth-invalid`, `not-entitled`,
103
- `user-blocked`, `app-unknown`, `rate-limited`, `upstream-busy`,
104
- `payload-too-large`, `modality-not-allowed`, `config-unavailable`,
105
- `upstream-error`, `internal`.
106
- - **Copy keys**: `ai.*` defaults in `aiMessages` — map through your i18n.
107
-
108
- ## Public API
109
-
110
- Everything is exported from `@chipmobilesdk/rn-ai`; internal `config/`,
111
- `transport/`, `core/`, `react/`, and `diagnostics/` paths are not public.
112
-
113
- - Client: `createAiClient`, `AiClient`, `AiClientConfig`, `TokenProvider`
114
- - Chat: `ChatRequest`, `ChatMessage`, `ContentPart`, `ChatStreamHandlers`,
115
- `ChatHandle`, `Usage`
116
- - Behavior and cache: `AppBehaviorConfig`, `AiBehaviorState`, `BehaviorState`,
117
- `AiConfigCache`, `QuotaSnapshot`
118
- - React: `useAiChat`, `useAiQuota`, `AiChatStatus`, `UseAiChatResult`,
119
- `UseAiQuotaResult`
120
- - Diagnostics and copy: `AiError`, `AiErrorCode`, `isAiError`, `aiMessages`,
121
- `aiMessage`, `AiMessageKey`
122
-
123
- ## App responsibilities
124
-
125
- - **AI disclosure (store policy)**: user text/images/audio are processed by
126
- the platform's AI service. Update your privacy policy, Apple App Privacy,
127
- and Google Play Data Safety before shipping, and disclose AI processing
128
- in-app (POL-005/006 in the feature spec).
129
- - Camera/microphone permissions for capture UIs are yours; the SDK only
130
- accepts already-captured data and adds no permissions.
131
- - Respect `payload-too-large` (default cap 10 MB/request) by resizing images
132
- and bounding audio length.
133
-
134
- ## Compatibility
135
-
136
- React ≥19 · React Native ≥0.85 (Hermes) · iOS + Android · Node ≥22.11 for
137
- tooling.
138
-
139
- The repository contains `src/screens/AiDemoScreen.tsx` and its component test.
140
- The current demo sidebar does not expose that screen yet, so it must be wired
141
- into `ScreenKey`, `Sidebar`, and `ActiveScreen` before it is reachable in-app.
142
-
143
- ## Validation
144
-
145
- Run from the repository root:
146
-
147
- ```bash
148
- npm run typecheck:ai
149
- npm run pack:ai
150
- npm run validate:ai
151
- ```
1
+ # @chipmobilesdk/rn-ai
2
+
3
+ AI gateway client for ChipMobileSdk-family React Native apps. Streams
4
+ multimodal (text + image + audio in, text out) AI responses through the
5
+ platform gateway with per-user rate limiting — apps never talk to Azure
6
+ OpenAI directly and hold no AI credentials.
7
+
8
+ > **License:** `UNLICENSED`. Published publicly for use by the owner's
9
+ > applications; no open-source license is granted.
10
+
11
+ - **Version**: see [CHANGELOG](CHANGELOG.md)
12
+ - **Entrypoint**: `@chipmobilesdk/rn-ai`
13
+ - **Native code**: none
14
+
15
+ - Contract: [rn-ai-public-api](../../specs/005-azure-ai-gateway/contracts/rn-ai-public-api.md)
16
+ - Gateway API: [gateway-api](../../specs/005-azure-ai-gateway/contracts/gateway-api.md)
17
+ - Behavior config: [app-behavior-config](../../specs/005-azure-ai-gateway/contracts/app-behavior-config.md)
18
+
19
+ ## Install
20
+
21
+ ```bash
22
+ npm install @chipmobilesdk/rn-ai @chipmobilesdk/rn-auth
23
+ ```
24
+
25
+ Peers: `react >=19`, `react-native >=0.85`, `@chipmobilesdk/rn-auth >=0.2.0`.
26
+ **Zero runtime dependencies**; streaming uses React Native's standard
27
+ `XMLHttpRequest` (no native code, Hermes-compatible, iOS + Android).
28
+
29
+ ## Setup
30
+
31
+ 1. Add the gateway resource to your rn-auth config:
32
+
33
+ ```ts
34
+ resources: [
35
+ { key: 'ai-gateway', scopes: ['api://aiapp-gateway/AI.Access'] },
36
+ ]
37
+ ```
38
+
39
+ 2. Create the client (one per app):
40
+
41
+ ```ts
42
+ import { createAiClient } from '@chipmobilesdk/rn-ai';
43
+ import { auth } from './auth/setup';
44
+
45
+ export const aiClient = createAiClient({
46
+ appId: 'scan-food',
47
+ configUrl: 'https://<config-host>/appconfig/config/scan-food.json',
48
+ resourceKey: 'ai-gateway',
49
+ auth,
50
+ // storage: inject a persistent AiConfigCache for offline launches
51
+ });
52
+ ```
53
+
54
+ 3. Use the hooks:
55
+
56
+ ```tsx
57
+ import { useAiChat, useAiQuota } from '@chipmobilesdk/rn-ai';
58
+ import { aiClient } from './ai/setup';
59
+
60
+ const { send, cancel, status, text, error, retryAt } = useAiChat(aiClient);
61
+ const { quota } = useAiQuota(aiClient);
62
+
63
+ send({
64
+ messages: [{
65
+ role: 'user',
66
+ content: [
67
+ { type: 'text', text: 'What dish is this? Estimate calories.' },
68
+ { type: 'image', mimeType: 'image/jpeg', dataBase64: photoBase64 },
69
+ ],
70
+ }],
71
+ });
72
+ ```
73
+
74
+ ## Runtime behavior
75
+
76
+ - **Config-driven endpoint**: resolved lazily on the first quota or chat
77
+ request. A fresh cache entry is used immediately; a stale entry is used
78
+ immediately and revalidated in the background with its ETag. With no cache,
79
+ the first operation waits for the config request, for at most `configMs`.
80
+ Concurrent calls share one config request, so a cold start that asks for a
81
+ chat and the quota together fetches the config once. Designated `404`/`410`
82
+ responses invalidate config and retry against the refreshed endpoint, after
83
+ waiting `recovery.retryBackoffMs[n]` (the last step repeats, each capped at
84
+ 30 s), at most `recovery.maxSilentRetries` times. A chat cancelled during the
85
+ wait is not retried.
86
+ - **Auth**: acquires tokens via rn-auth (`requestAccessToken('ai-gateway')`);
87
+ on `401`, reacquires a token and retries once.
88
+ - **Timeouts**: every network call is bounded. Override any of them with
89
+ `createAiClient({ timeouts: { ... } })`:
90
+
91
+ | | Default | Fails with |
92
+ |---|---|---|
93
+ | `connectMs` — chat request sent → headers | 30 s | `upstream-error`, `retryable: true` |
94
+ | `idleMs` — silence between two stream chunks | 30 s | `upstream-error`, `retryable: true` |
95
+ | `configMs` — behavior-config fetch | 10 s | stale cache if any, else `config-unavailable` |
96
+ | `quotaMs` — quota fetch | 10 s | `upstream-error`, `retryable: true` |
97
+
98
+ The gateway sends headers only after the model answers, so `connectMs` also
99
+ covers time to first token. A dropped connection (HTTP status 0) is likewise
100
+ `upstream-error` with `retryable: true`.
101
+ - **Callbacks**: each `chat()` delivers exactly one of `onDone` / `onError`,
102
+ and nothing after `cancel()`. An unexpected exception anywhere on the way
103
+ (including in your own `onDelta`) is reported as `internal`, never dropped.
104
+ - **Rate limits**: `429` surfaces as `status: 'rate-limited'` with `retryAt`;
105
+ the SDK never hammers and never enforces limits locally (display only).
106
+ - **Typed errors** (`AiError.code`): `auth-invalid`, `not-entitled`,
107
+ `user-blocked`, `app-unknown`, `rate-limited`, `upstream-busy`,
108
+ `payload-too-large`, `modality-not-allowed`, `config-unavailable`,
109
+ `upstream-error`, `internal`.
110
+ - **Copy keys**: `ai.*` defaults in `aiMessages` — map through your i18n.
111
+
112
+ ## Public API
113
+
114
+ Everything is exported from `@chipmobilesdk/rn-ai`; internal `config/`,
115
+ `transport/`, `core/`, `react/`, and `diagnostics/` paths are not public.
116
+
117
+ - Client: `createAiClient`, `AiClient`, `AiClientConfig`, `TokenProvider`
118
+ - Chat: `ChatRequest`, `ChatMessage`, `ContentPart`, `ChatStreamHandlers`,
119
+ `ChatHandle`, `Usage`
120
+ - Behavior and cache: `AppBehaviorConfig`, `AiBehaviorState`, `BehaviorState`,
121
+ `AiConfigCache`, `QuotaSnapshot`
122
+ - React: `useAiChat`, `useAiQuota`, `useAiBehaviorState`, `AiChatStatus`,
123
+ `UseAiChatResult`, `UseAiQuotaResult`. `useAiBehaviorState(client)` re-renders when
124
+ the config loads or the `aiEnabled` kill switch flips; `client.subscribe` and
125
+ `client.getBehaviorState()` give the same outside React.
126
+ - Diagnostics and copy: `AiError`, `AiErrorCode`, `isAiError`, `aiMessages`,
127
+ `aiMessage`, `AiMessageKey`, `aiMessageSet` (the same catalog as an rn-core message
128
+ set), `messageKeyForError(error)` (the copy to show for any error code). Register the
129
+ copy with rn-i18n in one line: `runtimeEntriesFromCatalog(aiMessageSet.user)`.
130
+
131
+ ## App responsibilities
132
+
133
+ - **AI disclosure (store policy)**: user text/images/audio are processed by
134
+ the platform's AI service. Update your privacy policy, Apple App Privacy,
135
+ and Google Play Data Safety before shipping, and disclose AI processing
136
+ in-app (POL-005/006 in the feature spec).
137
+ - Camera/microphone permissions for capture UIs are yours; the SDK only
138
+ accepts already-captured data and adds no permissions.
139
+ - Respect `payload-too-large` (default cap 10 MB/request) by resizing images
140
+ and bounding audio length.
141
+
142
+ ## Compatibility
143
+
144
+ React ≥19 · React Native ≥0.85 (Hermes) · iOS + Android · Node ≥22.11 for
145
+ tooling.
146
+
147
+ The repository contains `src/screens/AiDemoScreen.tsx` and its component test.
148
+ The current demo sidebar does not expose that screen yet, so it must be wired
149
+ into `ScreenKey`, `Sidebar`, and `ActiveScreen` before it is reachable in-app.
150
+
151
+ ## Validation
152
+
153
+ Run from the repository root:
154
+
155
+ ```bash
156
+ npm run typecheck --workspace @chipmobilesdk/rn-ai
157
+ npm run pack:check --workspace @chipmobilesdk/rn-ai
158
+ npm run validate
159
+ ```
package/package.json CHANGED
@@ -1,43 +1,39 @@
1
- {
2
- "name": "@chipmobilesdk/rn-ai",
3
- "version": "0.2.0",
4
- "private": false,
5
- "license": "UNLICENSED",
6
- "main": "src/index.ts",
7
- "types": "src/index.ts",
8
- "exports": {
9
- ".": {
10
- "types": "./src/index.ts",
11
- "react-native": "./src/index.ts",
12
- "default": "./src/index.ts"
13
- }
14
- },
15
- "files": [
16
- "src/index.ts",
17
- "src/config",
18
- "src/transport",
19
- "src/core",
20
- "src/react",
21
- "src/diagnostics",
22
- "skills",
23
- "README.md",
24
- "CHANGELOG.md"
25
- ],
26
- "sideEffects": false,
27
- "scripts": {
28
- "typecheck": "tsc -p tsconfig.json --noEmit",
29
- "pack:dry-run": "npm pack --dry-run"
30
- },
31
- "peerDependencies": {
32
- "@chipmobilesdk/rn-auth": ">=0.2.0",
33
- "react": ">=19",
34
- "react-native": ">=0.85"
35
- },
36
- "engines": {
37
- "node": ">=22.11.0"
38
- },
39
- "publishConfig": {
40
- "access": "public",
41
- "registry": "https://registry.npmjs.org/"
42
- }
43
- }
1
+ {
2
+ "name": "@chipmobilesdk/rn-ai",
3
+ "version": "0.3.0",
4
+ "private": false,
5
+ "license": "UNLICENSED",
6
+ "main": "src/index.ts",
7
+ "types": "src/index.ts",
8
+ "exports": {
9
+ ".": {
10
+ "types": "./src/index.ts",
11
+ "react-native": "./src/index.ts",
12
+ "default": "./src/index.ts"
13
+ }
14
+ },
15
+ "files": [
16
+ "src",
17
+ "!src/**/__tests__",
18
+ "skills",
19
+ "README.md",
20
+ "CHANGELOG.md"
21
+ ],
22
+ "sideEffects": false,
23
+ "scripts": {
24
+ "typecheck": "tsc -p tsconfig.json && tsc -p tsconfig.test.json",
25
+ "pack:check": "npm pack --dry-run"
26
+ },
27
+ "dependencies": {
28
+ "@chipmobilesdk/rn-core": "^0.1.0"
29
+ },
30
+ "peerDependencies": {
31
+ "@chipmobilesdk/rn-auth": ">=0.2.0",
32
+ "react": ">=19",
33
+ "react-native": ">=0.85"
34
+ },
35
+ "publishConfig": {
36
+ "access": "public",
37
+ "registry": "https://registry.npmjs.org/"
38
+ }
39
+ }