@opengeni/config 1.4.2 → 1.4.4

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/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@opengeni/config",
3
- "version": "1.4.2",
3
+ "version": "1.4.4",
4
4
  "description": "OpenGeni runtime configuration: settings resolution, deployment knobs, and config validation shared across the server packages.",
5
5
  "license": "Apache-2.0",
6
6
  "repository": {
@@ -33,9 +33,9 @@
33
33
  "prepublishOnly": "bash ../../scripts/prepublish-guard"
34
34
  },
35
35
  "dependencies": {
36
- "@opengeni/codex": "1.4.2",
37
- "@opengeni/contracts": "1.4.2",
38
- "@opengeni/xai-subscription": "1.4.2",
36
+ "@opengeni/codex": "1.4.4",
37
+ "@opengeni/contracts": "1.4.4",
38
+ "@opengeni/xai-subscription": "1.4.4",
39
39
  "zod": "^4.2.1"
40
40
  },
41
41
  "engines": {
package/src/index.ts CHANGED
@@ -67,6 +67,24 @@ import { createHash } from "node:crypto";
67
67
  import { isIP } from "node:net";
68
68
  import { z } from "zod";
69
69
 
70
+ /**
71
+ * Default maximum provider silence on a generic OpenAI-compatible model stream
72
+ * while the worker is waiting for the next chunk. Matches the Codex transport
73
+ * and SuperGrok idle defaults (5 minutes): long enough for a reasoning model's
74
+ * silent thinking gaps, short enough that a wedged stream recovers the same
75
+ * turn instead of hanging until a worker roll.
76
+ */
77
+ export const DEFAULT_MODEL_STREAM_IDLE_TIMEOUT_MS = 300_000;
78
+ /**
79
+ * Default maximum wait across keepalive-only traffic (SSE comments, `ping` /
80
+ * `keepalive` events, repeated `response.in_progress`) for the next
81
+ * model-progress event on a generic model stream. A provider can keep a
82
+ * connection alive while its generation is wedged; this bounds that case.
83
+ * Ten minutes leaves room for a reasoning model's longest silent thinking
84
+ * window (reasoning summaries stream as progress).
85
+ */
86
+ export const DEFAULT_MODEL_STREAM_PROGRESS_TIMEOUT_MS = 600_000;
87
+
70
88
  const envName = /^[A-Za-z_][A-Za-z0-9_]*$/;
71
89
  const registryId = /^[A-Za-z0-9_-]+$/;
72
90
  export const DEFAULT_OPENROUTER_MODEL_ID =
@@ -809,6 +827,29 @@ const SettingsSchema = z.object({
809
827
  .positive()
810
828
  .max(24 * 60 * 60_000)
811
829
  .default(XAI_RESPONSE_STREAM_IDLE_TIMEOUT_MS),
830
+ // Maximum silence between response bytes on a generic OpenAI-compatible
831
+ // model stream (built-in OpenAI/Azure and registry chat/responses providers).
832
+ // Not a request/run duration cap: every received byte resets the timer, and
833
+ // it measures only time spent waiting on the provider for the next chunk. A
834
+ // stalled stream fails as a retryable provider failure so finite same-turn
835
+ // recovery replaces it. A registry provider may override it with
836
+ // `streamIdleTimeoutMs`. Codex, SuperGrok, and Anthropic Messages own their
837
+ // transport-specific idle timers.
838
+ modelStreamIdleTimeoutMs: z.coerce
839
+ .number()
840
+ .int()
841
+ .positive()
842
+ .max(24 * 60 * 60_000)
843
+ .default(DEFAULT_MODEL_STREAM_IDLE_TIMEOUT_MS),
844
+ // Maximum wait across keepalive-only stream traffic for the next
845
+ // model-progress event (registry override: `streamProgressTimeoutMs`).
846
+ // Never shorter than modelStreamIdleTimeoutMs.
847
+ modelStreamProgressTimeoutMs: z.coerce
848
+ .number()
849
+ .int()
850
+ .positive()
851
+ .max(24 * 60 * 60_000)
852
+ .default(DEFAULT_MODEL_STREAM_PROGRESS_TIMEOUT_MS),
812
853
  // Expose the connected apps attached to a Codex subscription through the
813
854
  // synthetic codex_apps MCP server. Independent from subscription routing so
814
855
  // operators can use Codex models without exposing ChatGPT connectors.
@@ -2682,6 +2723,20 @@ const RegistryProviderSchema = z
2682
2723
  label: z.string().min(1).optional(),
2683
2724
  api: ModelProviderApi.default("chat"),
2684
2725
  anthropic: AnthropicProviderOptions.optional(),
2726
+ // Generic OpenAI-compatible stream silence bound; overrides the deployment
2727
+ // `modelStreamIdleTimeoutMs` for this provider only.
2728
+ streamIdleTimeoutMs: z
2729
+ .number()
2730
+ .int()
2731
+ .positive()
2732
+ .max(24 * 60 * 60_000)
2733
+ .optional(),
2734
+ streamProgressTimeoutMs: z
2735
+ .number()
2736
+ .int()
2737
+ .positive()
2738
+ .max(24 * 60 * 60_000)
2739
+ .optional(),
2685
2740
  wireProfile: ModelProviderWireProfile.default("openai"),
2686
2741
  baseUrl: z.string().url(),
2687
2742
  apiKey: z.string().optional(), // inline key (pragmatic) ...
@@ -3174,6 +3229,10 @@ export interface ResolvedModelProvider {
3174
3229
  defaultHeaders?: Record<string, string> | undefined;
3175
3230
  publicDefaultQueryNames?: string[] | undefined;
3176
3231
  publicDefaultHeaderNames?: string[] | undefined;
3232
+ /** Generic OpenAI-compatible stream silence bound override (ms). */
3233
+ streamIdleTimeoutMs?: number | undefined;
3234
+ /** Generic OpenAI-compatible keepalive-only progress bound override (ms). */
3235
+ streamProgressTimeoutMs?: number | undefined;
3177
3236
  credentialSource: CredentialSourceV1;
3178
3237
  billing: BillingAttributionV1;
3179
3238
  }
@@ -4049,6 +4108,8 @@ export function getSettings(source: NodeJS.ProcessEnv = process.env): Settings {
4049
4108
  supergrokResponseStreamIdleTimeoutMs: optional(
4050
4109
  "OPENGENI_SUPERGROK_RESPONSE_STREAM_IDLE_TIMEOUT_MS",
4051
4110
  ),
4111
+ modelStreamIdleTimeoutMs: optional("OPENGENI_MODEL_STREAM_IDLE_TIMEOUT_MS"),
4112
+ modelStreamProgressTimeoutMs: optional("OPENGENI_MODEL_STREAM_PROGRESS_TIMEOUT_MS"),
4052
4113
  codexConnectedAppsEnabled: optional("OPENGENI_CODEX_CONNECTED_APPS_ENABLED"),
4053
4114
  codexToolSearchEnabled: optional("OPENGENI_CODEX_TOOL_SEARCH_ENABLED"),
4054
4115
  lazyToolSearchEnabled: optional("OPENGENI_LAZY_TOOL_SEARCH_ENABLED"),
@@ -6006,6 +6067,12 @@ export function configuredProviders(
6006
6067
  defaultHeaders: provider.defaultHeaders,
6007
6068
  publicDefaultQueryNames: provider.publicDefaultQueryNames,
6008
6069
  publicDefaultHeaderNames: provider.publicDefaultHeaderNames,
6070
+ ...(provider.streamIdleTimeoutMs !== undefined
6071
+ ? { streamIdleTimeoutMs: provider.streamIdleTimeoutMs }
6072
+ : {}),
6073
+ ...(provider.streamProgressTimeoutMs !== undefined
6074
+ ? { streamProgressTimeoutMs: provider.streamProgressTimeoutMs }
6075
+ : {}),
6009
6076
  credentialSource: registryCredentialSource(provider),
6010
6077
  billing: registryBilling(provider),
6011
6078
  }),