@linxiraos/pi-catalog 1.4.4 → 1.4.7

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.
Files changed (103) hide show
  1. package/CHANGELOG.md +31 -0
  2. package/README.md +3 -3
  3. package/dist/types/build.d.ts +12 -4
  4. package/dist/types/compat/anthropic.d.ts +20 -0
  5. package/dist/types/compat/behavior.d.ts +2 -0
  6. package/dist/types/compat/catalog-policy.d.ts +2 -2
  7. package/dist/types/compat/collapse.d.ts +10 -0
  8. package/dist/types/compat/factory-droid.d.ts +2 -0
  9. package/dist/types/compat/resolve.d.ts +2 -2
  10. package/dist/types/compat/types.d.ts +5 -1
  11. package/dist/types/discovery/antigravity.d.ts +7 -2
  12. package/dist/types/discovery/codex.d.ts +4 -5
  13. package/dist/types/discovery/devin.d.ts +1 -1
  14. package/dist/types/model-cache.d.ts +8 -0
  15. package/dist/types/model-catalog-format.d.ts +37 -0
  16. package/dist/types/models.d.ts +6 -3
  17. package/dist/types/provider-models/google.d.ts +2 -0
  18. package/dist/types/provider-models/openai-compat.d.ts +16 -3
  19. package/dist/types/types.d.ts +58 -3
  20. package/dist/types/wire/codex.d.ts +1 -1
  21. package/dist/types/wire/devin.d.ts +2 -2
  22. package/dist/types/wire/gemini-headers.d.ts +41 -22
  23. package/dist/types/wire/xiaomi-token-plan.d.ts +2 -0
  24. package/package.json +6 -10
  25. package/src/build.d.ts +12 -4
  26. package/src/build.ts +97 -4
  27. package/src/compat/anthropic.d.ts +20 -0
  28. package/src/compat/anthropic.ts +47 -0
  29. package/src/compat/axes.ts +16 -3
  30. package/src/compat/behavior.d.ts +2 -0
  31. package/src/compat/behavior.ts +5 -0
  32. package/src/compat/catalog-policy.ts +10 -8
  33. package/src/compat/collapse.d.ts +10 -0
  34. package/src/compat/collapse.ts +63 -2
  35. package/src/compat/factory-droid.d.ts +2 -0
  36. package/src/compat/factory-droid.ts +3 -0
  37. package/src/compat/resolve.d.ts +2 -2
  38. package/src/compat/resolve.ts +7 -5
  39. package/src/compat/rules/README.md +24 -16
  40. package/src/compat/rules/auth/google-antigravity.kdl +2 -1
  41. package/src/compat/rules/auth/llama.cpp.kdl +1 -1
  42. package/src/compat/rules/auth/typesafe.kdl +2 -1
  43. package/src/compat/rules/auth/xiaomi-token-plan-ams.kdl +1 -1
  44. package/src/compat/rules/auth/xiaomi-token-plan-cn.kdl +1 -1
  45. package/src/compat/rules/auth/xiaomi-token-plan-sgp.kdl +1 -1
  46. package/src/compat/rules/classes/anthropic.kdl +6 -1
  47. package/src/compat/rules/classes/deepseek.kdl +2 -1
  48. package/src/compat/rules/classes/minimax.kdl +10 -0
  49. package/src/compat/rules/classes/openai.kdl +2 -2
  50. package/src/compat/rules/providers/anthropic.kdl +15 -1
  51. package/src/compat/rules/providers/deepseek.kdl +13 -10
  52. package/src/compat/rules/providers/factory-droid.kdl +72 -0
  53. package/src/compat/rules/providers/google-antigravity.kdl +7 -0
  54. package/src/compat/rules/providers/google-vertex.kdl +3 -0
  55. package/src/compat/rules/providers/kimi-code.kdl +26 -0
  56. package/src/compat/rules/providers/mistral.kdl +8 -0
  57. package/src/compat/rules/providers/moonshot.kdl +25 -0
  58. package/src/compat/rules/providers/muse-code.kdl +1 -1
  59. package/src/compat/rules/providers/openai-codex.kdl +17 -4
  60. package/src/compat/rules/providers/openai.kdl +23 -2
  61. package/src/compat/rules/providers/opencode-go.kdl +38 -3
  62. package/src/compat/rules/providers/opencode-zen.kdl +7 -0
  63. package/src/compat/rules/providers/xiaomi-token-plan-ams.kdl +1 -1
  64. package/src/compat/rules/providers/xiaomi-token-plan-cn.kdl +1 -1
  65. package/src/compat/rules/providers/xiaomi-token-plan-sgp.kdl +1 -1
  66. package/src/compat/rules/providers/xiaomi.kdl +1 -1
  67. package/src/compat/rules/runtime/behavior.kdl +39 -20
  68. package/src/compat/rules/taxonomy/kimi.kdl +7 -0
  69. package/src/compat/rules.json +1 -1
  70. package/src/compat/types.d.ts +5 -1
  71. package/src/compat/types.ts +5 -1
  72. package/src/discovery/antigravity.d.ts +7 -2
  73. package/src/discovery/antigravity.ts +41 -6
  74. package/src/discovery/codex.d.ts +4 -5
  75. package/src/discovery/codex.ts +5 -6
  76. package/src/discovery/cursor.ts +0 -6
  77. package/src/discovery/devin.d.ts +1 -1
  78. package/src/discovery/devin.ts +9 -3
  79. package/src/discovery/factory-droid.ts +38 -6
  80. package/src/model-cache.d.ts +8 -0
  81. package/src/model-cache.ts +55 -5
  82. package/src/model-catalog-format.d.ts +37 -0
  83. package/src/model-catalog-format.ts +299 -0
  84. package/src/models.d.ts +6 -3
  85. package/src/models.json +1 -1
  86. package/src/models.ts +16 -10
  87. package/src/provider-models/cache-provider-id.ts +26 -6
  88. package/src/provider-models/google.d.ts +2 -0
  89. package/src/provider-models/google.ts +13 -1
  90. package/src/provider-models/openai-compat.d.ts +16 -3
  91. package/src/provider-models/openai-compat.ts +207 -58
  92. package/src/provider-models/special.ts +5 -6
  93. package/src/types.d.ts +58 -3
  94. package/src/types.ts +63 -1
  95. package/src/wire/codex.d.ts +1 -1
  96. package/src/wire/codex.ts +1 -1
  97. package/src/wire/devin.d.ts +2 -2
  98. package/src/wire/devin.ts +2 -2
  99. package/src/wire/gemini-headers.d.ts +41 -22
  100. package/src/wire/gemini-headers.ts +68 -24
  101. package/src/wire/xiaomi-token-plan.d.ts +2 -0
  102. package/src/wire/xiaomi-token-plan.ts +4 -0
  103. package/src/models.json.d.ts +0 -9
package/src/models.ts CHANGED
@@ -1,5 +1,6 @@
1
1
  import { classifyModel } from "./compat/taxonomy";
2
- import MODELS from "./models.json" with { type: "json" };
2
+ import { getCatalogProvider, getCatalogProviders, openModelCatalog } from "./model-catalog-format";
3
+ import modelsData from "./models.json" with { type: "json" };
3
4
  import { dueEffectiveRate } from "./pricing";
4
5
  import type {
5
6
  Api,
@@ -21,20 +22,21 @@ import type {
21
22
  *
22
23
  * For runtime-aware resolution, use `createModelManager()` / `resolveProviderModels()`.
23
24
  */
25
+ const catalog = openModelCatalog(modelsData);
26
+ const bundledProviders: readonly KnownProvider[] = Object.freeze(getCatalogProviders(catalog) as KnownProvider[]);
27
+ const bundledProviderSet = new Set<string>(bundledProviders);
24
28
  const modelRegistry = new Map<string, Map<string, Model<Api>>>();
25
29
 
26
30
  /** Return one provider's bundled models, materialized by the generator. */
27
31
  function getProviderModels(provider: string): Map<string, Model<Api>> | undefined {
28
32
  const cachedModels = modelRegistry.get(provider);
29
33
  if (cachedModels !== undefined) return cachedModels;
30
- if (!Object.hasOwn(MODELS, provider)) return undefined;
34
+ const rawModels = getCatalogProvider(catalog, provider);
35
+ if (!rawModels) return undefined;
31
36
 
32
37
  const providerModels = new Map<string, Model<Api>>();
33
- const rawModels = MODELS[provider as keyof typeof MODELS];
34
38
  for (const id in rawModels) {
35
- // models.json rows are complete Models emitted by generate-models.ts;
36
- // consuming them verbatim keeps startup allocation-free.
37
- const row = rawModels[id as keyof typeof rawModels] as unknown as Model<Api>;
39
+ const row = rawModels[id];
38
40
  // Rows baked before the compat engine (and stale cache snapshots) lack
39
41
  // `identity`; classify lazily so consumers can rely on the field.
40
42
  row.identity ??= classifyModel(provider, id, { lenient: true });
@@ -44,22 +46,26 @@ function getProviderModels(provider: string): Map<string, Model<Api>> | undefine
44
46
  return providerModels;
45
47
  }
46
48
 
47
- export type GeneratedProvider = keyof typeof MODELS;
49
+ /** Provider identifier accepted by the bundled model registry. */
50
+ export type GeneratedProvider = string;
48
51
 
49
52
  /** Whether `provider` has bundled rows in models.json (e.g. a provider id authored in KDL). */
50
53
  export function isGeneratedProvider(provider: string): provider is GeneratedProvider {
51
- return Object.hasOwn(MODELS, provider);
54
+ return bundledProviderSet.has(provider);
52
55
  }
53
56
 
57
+ /** Look up a bundled model; nested catalog values are shared and must not be mutated. */
54
58
  export function getBundledModel<TApi extends Api = Api>(provider: GeneratedProvider, modelId: string): Model<TApi> {
55
59
  const providerModels = getProviderModels(provider);
56
60
  return providerModels?.get(modelId) as Model<TApi>;
57
61
  }
58
62
 
59
- export function getBundledProviders(): KnownProvider[] {
60
- return Object.keys(MODELS) as KnownProvider[];
63
+ /** List providers with bundled rows; the array is shared and frozen. */
64
+ export function getBundledProviders(): readonly KnownProvider[] {
65
+ return bundledProviders;
61
66
  }
62
67
 
68
+ /** List a provider's bundled models; nested catalog values are shared and must not be mutated. */
63
69
  export function getBundledModels(provider: GeneratedProvider): Model<Api>[] {
64
70
  const models = getProviderModels(provider);
65
71
  return models ? (Array.from(models.values()) as Model<Api>[]) : [];
@@ -26,6 +26,9 @@ const CREDENTIAL_SCOPED_MODEL_CACHE_PROVIDERS: Readonly<Record<string, true>> =
26
26
  // than from the synchronous, credential-less startup read.
27
27
  "singularityapi-dev": true,
28
28
  "singularityapi-tech": true,
29
+ // Kimi Code's roster is plan-gated and discovery is authoritative, so a
30
+ // shared namespace would serve one plan's ids to another plan's credential.
31
+ "kimi-code": true,
29
32
  };
30
33
 
31
34
  /** Whether a provider's model-cache namespace requires its resolved credential. */
@@ -37,6 +40,8 @@ export function getDefaultModelDiscoveryBaseUrl(providerId: string): string | un
37
40
  switch (providerId) {
38
41
  case "charm-hyper":
39
42
  return CHARM_HYPER_API_BASE_URL;
43
+ case "kimi-code":
44
+ return "https://api.kimi.com/coding/v1";
40
45
  case "meta":
41
46
  case "muse-code":
42
47
  return "https://api.meta.ai/v1";
@@ -71,11 +76,11 @@ export function resolveOllamaModelCacheProviderId(providerId: string, baseUrl?:
71
76
  }
72
77
 
73
78
  /**
74
- * The `sub` claim of a Cursor access token. Cursor access tokens are JWTs
75
- * that `refreshCursorToken` rotates (new `exp`/`iat`) for the same account,
79
+ * The `sub` claim of a JWT access token. Cursor and Kimi Code access tokens
80
+ * are JWTs that refresh rotates (new `exp`/`iat`/`jti`) for the same account,
76
81
  * so the subject is the stable account scope. Non-JWT keys return undefined.
77
82
  */
78
- function cursorCredentialSubject(apiKey: string): string | undefined {
83
+ function jwtCredentialSubject(apiKey: string): string | undefined {
79
84
  const parts = apiKey.split(".");
80
85
  if (parts.length !== 3 || !parts[1]) return undefined;
81
86
  try {
@@ -112,7 +117,7 @@ export function resolveModelCacheProviderId(providerId: string, options: ModelCa
112
117
  // mark the account default `isProviderDefault`, which startup now honors.
113
118
  const baseUrl = (options.baseUrl ?? CURSOR_DEFAULT_BASE_URL).replace(/\/+$/, "");
114
119
  const apiKey = options.apiKey ?? "";
115
- const scope = `${cursorCredentialSubject(apiKey) ?? apiKey}\u0000${baseUrl}`;
120
+ const scope = `${jwtCredentialSubject(apiKey) ?? apiKey}\u0000${baseUrl}`;
116
121
  return `cursor:rich-models-v5:${Bun.hash(scope).toString(36)}`;
117
122
  }
118
123
  case "charm-hyper": {
@@ -132,6 +137,17 @@ export function resolveModelCacheProviderId(providerId: string, options: ModelCa
132
137
  // `/v1`-suffixed one; both must land on one namespace.
133
138
  return `charm-hyper:models-v1:${Bun.hash(normalizeCharmHyperBaseUrl(options.baseUrl)).toString(36)}`;
134
139
  }
140
+ case "kimi-code": {
141
+ // The `/coding/v1/models` roster depends on the subscription plan behind
142
+ // the credential, and discovery is authoritative: a namespace without the
143
+ // credential would serve the previous plan's roster for the full TTL,
144
+ // including ids the current plan cannot call. OAuth access tokens rotate on
145
+ // refresh, so the scope is the token's account subject, not the token.
146
+ const baseUrl = options.baseUrl ?? getDefaultModelDiscoveryBaseUrl(providerId)!;
147
+ const apiKey = options.apiKey ?? "";
148
+ const scope = `${jwtCredentialSubject(apiKey) ?? apiKey}\u0000${baseUrl}`;
149
+ return `kimi-code:models-v1:${Bun.hash(scope).toString(36)}`;
150
+ }
135
151
  case "muse-code": {
136
152
  const baseUrl = options.baseUrl ?? getDefaultModelDiscoveryBaseUrl(providerId)!;
137
153
  const scope = `${options.apiKey ?? ""}\u0000${baseUrl}`;
@@ -160,6 +176,8 @@ export function resolveModelCacheProviderId(providerId: string, options: ModelCa
160
176
  }
161
177
  case "litellm": {
162
178
  const baseUrl = options.baseUrl ?? getDefaultModelDiscoveryBaseUrl(providerId)!;
179
+ // rich-v14 fetches every /v2/model/info page rather than caching only
180
+ // its first 50 models. rich-v13 moves `mode: decisions` groups off chat.
163
181
  // rich-v12 invalidates namespaced ids that missed bare catalog references.
164
182
  // rich-v11 excluded ClinePass gateway metadata (issue #10932). rich-v10
165
183
  // filtered known non-conversational LiteLLM modes, unioned compat across
@@ -167,7 +185,7 @@ export function resolveModelCacheProviderId(providerId: string, options: ModelCa
167
185
  // declaration into it; earlier versions invalidated rows whose
168
186
  // `compatConfig` retained a colliding bundled model's provider-specific
169
187
  // transport (e.g. Fireworks `wireModelIdMode`) (issue #9938).
170
- return `litellm:rich-v12:${Bun.hash(baseUrl).toString(36)}`;
188
+ return `litellm:rich-v14:${Bun.hash(baseUrl).toString(36)}`;
171
189
  }
172
190
  case "gmi-cloud":
173
191
  case "siliconflow":
@@ -211,7 +229,9 @@ export function resolveModelCacheProviderId(providerId: string, options: ModelCa
211
229
  // v2: rows cached before Fusion pairings carried their lead uid as
212
230
  // `requestModelId` send the composite uid and fail with
213
231
  // `permission_denied: no API providers are available`.
214
- return "devin:models-v2";
232
+ // v3: rows cached before discovery read `ModelInfo.maxTokens` hold a
233
+ // 200K window on every model.
234
+ return "devin:models-v3";
215
235
  default:
216
236
  return providerId;
217
237
  }
@@ -21,6 +21,8 @@ export interface GoogleAntigravityAccount {
21
21
  * When any account lacks one, no model records per-account access.
22
22
  */
23
23
  accountKey?: string;
24
+ /** Cloud Code Assist project of the credential, sent with roster discovery like the native client. */
25
+ projectId?: string;
24
26
  }
25
27
  export interface GoogleAntigravityModelManagerConfig {
26
28
  /**
@@ -33,6 +33,8 @@ export interface GoogleAntigravityAccount {
33
33
  * When any account lacks one, no model records per-account access.
34
34
  */
35
35
  accountKey?: string;
36
+ /** Cloud Code Assist project of the credential, sent with roster discovery like the native client. */
37
+ projectId?: string;
36
38
  }
37
39
 
38
40
  export interface GoogleAntigravityModelManagerConfig {
@@ -113,6 +115,7 @@ export function googleAntigravityModelManagerOptions(
113
115
  result: await fetchAntigravityDiscoveryModels({
114
116
  token: account.accessToken,
115
117
  endpoint: config?.endpoint,
118
+ projectId: account.projectId,
116
119
  fetcher,
117
120
  }),
118
121
  })),
@@ -140,7 +143,15 @@ export function googleAntigravityModelManagerOptions(
140
143
  return unionAccountCatalogs(
141
144
  catalogs.map(({ accountKey, models }) =>
142
145
  tagAccess && accountKey !== undefined
143
- ? models.map(model => ({ ...model, accountAccess: { [accountKey]: {} } }))
146
+ ? models.map(model => ({
147
+ ...model,
148
+ accountAccess: {
149
+ [accountKey]:
150
+ model.antigravityWireProfiles === undefined
151
+ ? {}
152
+ : { antigravityWireProfiles: model.antigravityWireProfiles },
153
+ },
154
+ }))
144
155
  : models,
145
156
  ),
146
157
  );
@@ -167,6 +178,7 @@ export function googleGeminiCliModelManagerOptions(
167
178
  }
168
179
  const result = await fetchAntigravityDiscoveryModels({
169
180
  token,
181
+ projectId: config?.projectId,
170
182
  fetcher,
171
183
  collapseTable,
172
184
  });
@@ -377,24 +377,32 @@ export declare const KIMI_CODE_DEFAULT_MAX_TOKENS = 32000;
377
377
  export declare function kimiCodeMaxTokens(modelId: string, fallback?: number): number;
378
378
  export declare function kimiCodeMaxTokens(modelId: string, fallback: number | null): number | null;
379
379
  export declare function kimiCodeModelManagerOptions(config?: KimiCodeModelManagerConfig): ModelManagerOptions<"openai-completions">;
380
- /** Native LM Studio metadata keyed by model id from `/api/v0/models`. */
380
+ /** Native LM Studio metadata keyed by model id from `/api/v0/models` and `/api/v1/models`. */
381
381
  export interface LmStudioNativeModelMetadata {
382
382
  input: ("text" | "image")[];
383
383
  contextWindow?: number;
384
+ /**
385
+ * LM Studio reports a decision model (`type: "decision"`): it answers typed
386
+ * questions on `/v1/decisions` and rejects chat completions.
387
+ */
388
+ decision?: true;
384
389
  }
385
390
  /** Options for LM Studio's optional native metadata probe. */
386
391
  export interface LmStudioNativeModelMetadataOptions {
387
392
  headers?: Record<string, string>;
388
393
  signal?: AbortSignal;
389
394
  }
390
- /** Fetches LM Studio native model metadata used to mark VLM models as image-capable. */
395
+ /**
396
+ * Fetches LM Studio native model metadata used to mark VLM models as
397
+ * image-capable and decision models as judges.
398
+ */
391
399
  export declare function fetchLmStudioNativeModelMetadata(baseUrl: string, fetchImpl?: FetchImpl, options?: LmStudioNativeModelMetadataOptions): Promise<Map<string, LmStudioNativeModelMetadata> | null>;
392
400
  export interface LmStudioModelManagerConfig {
393
401
  apiKey?: string;
394
402
  baseUrl?: string;
395
403
  fetch?: FetchImpl;
396
404
  }
397
- export declare function lmStudioModelManagerOptions(config?: LmStudioModelManagerConfig): ModelManagerOptions<"openai-completions">;
405
+ export declare function lmStudioModelManagerOptions(config?: LmStudioModelManagerConfig): ModelManagerOptions<Api>;
398
406
  export interface SyntheticModelManagerConfig {
399
407
  apiKey?: string;
400
408
  baseUrl?: string;
@@ -560,6 +568,11 @@ export declare const OPENAI_COMPAT_DISCOVERY_DEFAULT_MAX_TOKENS = 32768;
560
568
  /** Exclude only known non-conversational modes; unknown and non-string modes remain selectable for aliases. */
561
569
  export declare function isSelectableLiteLLMModelMode(mode: unknown): boolean;
562
570
  export declare function normalizeLiteLLMManagementBaseUrl(baseUrl: string): string;
571
+ /**
572
+ * Transport for a LiteLLM model group: `mode: "decisions"` answers only on the
573
+ * proxy's `/v1/decisions` route, OpenAI-backed groups use Responses, and every
574
+ * other group uses `fallbackApi`.
575
+ */
563
576
  export declare function resolveLiteLLMApi(entry: Record<string, unknown> | undefined, id: string, fallbackApi?: Api): Api;
564
577
  export declare function fetchLiteLLMRichModels<TApi extends Api>(options: FetchLiteLLMRichModelsOptions<TApi>): Promise<ModelSpec<TApi>[] | null>;
565
578
  export declare function litellmModelManagerOptions(config?: LiteLLMModelManagerConfig): ModelManagerOptions<Api>;