@prismshadow/penguin-server 0.2.16 → 0.2.17

package/dist/model-catalog.js

@@ -0,0 +1,108 @@
+import {
+  DEEPSEEK_OFF_PEAK,
+  MODELSCOPE_PROVIDER_ID,
+  MODEL_CATALOG,
+  MODEL_PROVIDERS,
+  PENGUIN_GO_BASE_URL,
+  PENGUIN_GO_PROVIDER_ID,
+  PRESET_DEFAULT_MODEL,
+  QWEN_OFF_PEAK,
+  catalogEntryFor,
+  catalogGroupConnection,
+  catalogModelEntries,
+  clientEnvKey,
+  effectivePricing,
+  fixedClientType,
+  fixedClientTypeRefuses,
+  isAddableGroup,
+  isVendorGroup,
+  modelCalledKeyless,
+  modelEnvPreviewKey,
+  modelEnvironment,
+  modelHomepageUrl,
+  offPeakAt,
+  offPeakScheduledRefs,
+  presetModelEntries,
+  presetPromotions,
+  presetProviderTable,
+  providerCalledKeyless,
+  providerClientType,
+  providerEnvFallbackKey,
+  providerInfo,
+  retainedCatalogPricing,
+  unaddableModel,
+  unroutableVendorModel,
+  vendorClientType
+} from "./chunk-GGIINEEK.js";
+import "./chunk-B5IP37ZQ.js";
+
+// src/model-catalog/client-protocols.ts
+import { routedClientType } from "@prismshadow/mmsp/credentials";
+import { routedClientType as routedClientType2 } from "@prismshadow/mmsp/credentials";
+var CLIENT_PROTOCOLS = {
+  "openai-official": { path: "/responses", fastMode: "openai" },
+  "anthropic-official": { path: "/v1/messages", fastMode: "anthropic" },
+  "google-official": { path: "/v1beta/interactions", fastMode: "openai" },
+  "zai-official": { path: "/chat/completions" },
+  "moonshot-official": { path: "/chat/completions" },
+  "deepseek-official": { path: "/responses" },
+  "minimax-official": { path: "/responses", fastMode: "openai" },
+  "openai-responses": { path: "/responses", fastMode: "openai" },
+  "openai-chat": { path: "/chat/completions", fastMode: "openai" },
+  "openai-chat-vllm-adapter": { path: "/chat/completions", fastMode: "openai" },
+  "openai-embedding": { path: "/embeddings" },
+  "ant-messages": { path: "/v1/messages", fastMode: "anthropic" },
+  "google-genai": { path: "/v1beta/models" },
+  // An MMSP server: the client posts to {base}/stream. It forwards `fast_mode` to the server,
+  // whose upstream client decides, so the harness offers no toggle it cannot vouch for.
+  mmsp: { path: "/stream" }
+};
+function fastModeProtocol(modelId, clientType, baseUrl) {
+  const routed = routedClientType(modelId, clientType?.trim() ? clientType : null);
+  if (routed === "anthropic-official" && (baseUrl?.startsWith("bedrock://") || ["4-6", "sonnet-5-5", "haiku-5-5", "fable-5-1"].some(
+    (generation) => modelId.includes(generation)
+  ))) {
+    return void 0;
+  }
+  return routed === null ? void 0 : CLIENT_PROTOCOLS[routed]?.fastMode;
+}
+export {
+  CLIENT_PROTOCOLS,
+  DEEPSEEK_OFF_PEAK,
+  MODELSCOPE_PROVIDER_ID,
+  MODEL_CATALOG,
+  MODEL_PROVIDERS,
+  PENGUIN_GO_BASE_URL,
+  PENGUIN_GO_PROVIDER_ID,
+  PRESET_DEFAULT_MODEL,
+  QWEN_OFF_PEAK,
+  catalogEntryFor,
+  catalogGroupConnection,
+  catalogModelEntries,
+  clientEnvKey,
+  effectivePricing,
+  fastModeProtocol,
+  fixedClientType,
+  fixedClientTypeRefuses,
+  isAddableGroup,
+  isVendorGroup,
+  modelCalledKeyless,
+  modelEnvPreviewKey,
+  modelEnvironment,
+  modelHomepageUrl,
+  offPeakAt,
+  offPeakScheduledRefs,
+  presetModelEntries,
+  presetPromotions,
+  presetProviderTable,
+  providerCalledKeyless,
+  providerClientType,
+  providerEnvFallbackKey,
+  providerInfo,
+  retainedCatalogPricing,
+  routedClientType2 as routedClientType,
+  unaddableModel,
+  unroutableVendorModel,
+  vendorClientType
+};
+//# sourceMappingURL=model-catalog.js.map

package/dist/model-catalog.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/model-catalog/client-protocols.ts"],"sourcesContent":["/**\n * What the UI needs to know about MMSP's clients: which client an entry reaches (MMSP's own\n * routedClientType, passed through), the path that client appends to its base URL, and the\n * protocol that carries fast mode on it. MMSP's credentials module answers the routing; the\n * catalog keeps no copy of it and adds only what MMSP exports nothing for (CLIENT_PROTOCOLS).\n *\n * A leaf: it imports MMSP's browser-safe credentials module and nothing of the catalog, so the\n * catalog, the environment-key previews and the browser may all read it.\n */\nimport { routedClientType } from \"@prismshadow/mmsp/credentials\";\n\n// MMSP's routing, passed through for browser code that reads the catalog: which client an entry\n// reaches (its pinned client type, else its id's family), or null where nothing routes.\nexport { routedClientType } from \"@prismshadow/mmsp/credentials\";\n\n/**\n * The wire protocol that carries MMSP's `fast_mode` on a client: `\"openai\"` for the clients\n * that send `service_tier: \"priority\"` (the OpenAI protocols, and Google's Interactions API\n * and MiniMax alike), `\"anthropic\"` for those that send `speed: \"fast\"` plus the\n * `fast-mode-2026-02-01` beta header. The two differ in what the user must be warned about,\n * not just in wire shape.\n */\nexport type FastModeProtocol = \"openai\" | \"anthropic\";\n\n/**\n * What the UI shows about each MMSP client, which MMSP exports nothing for: the path it appends\n * to its base URL (the models page's endpoint hint) and the protocol that carries `fast_mode`\n * (absent = the client rejects it, or the harness cannot tell what it does with it). Keyed by\n * the canonical client type MMSP's routedClientType returns; the catalog test holds it to the\n * client types MMSP's credentials module knows.\n */\nexport const CLIENT_PROTOCOLS: Readonly<\n  Record<string, { path: string; fastMode?: FastModeProtocol }>\n> = {\n  \"openai-official\": { path: \"/responses\", fastMode: \"openai\" },\n  \"anthropic-official\": { path: \"/v1/messages\", fastMode: \"anthropic\" },\n  \"google-official\": { path: \"/v1beta/interactions\", fastMode: \"openai\" },\n  \"zai-official\": { path: \"/chat/completions\" },\n  \"moonshot-official\": { path: \"/chat/completions\" },\n  \"deepseek-official\": { path: \"/responses\" },\n  \"minimax-official\": { path: \"/responses\", fastMode: \"openai\" },\n  \"openai-responses\": { path: \"/responses\", fastMode: \"openai\" },\n  \"openai-chat\": { path: \"/chat/completions\", fastMode: \"openai\" },\n  \"openai-chat-vllm-adapter\": { path: \"/chat/completions\", fastMode: \"openai\" },\n  \"openai-embedding\": { path: \"/embeddings\" },\n  \"ant-messages\": { path: \"/v1/messages\", fastMode: \"anthropic\" },\n  \"google-genai\": { path: \"/v1beta/models\" },\n  // An MMSP server: the client posts to {base}/stream. It forwards `fast_mode` to the server,\n  // whose upstream client decides, so the harness offers no toggle it cannot vouch for.\n  mmsp: { path: \"/stream\" },\n};\n\n/**\n * Whether a model can carry fast mode at all, and on which protocol - `undefined` means no.\n *\n * The fast tier is a property of the **client MMSP routes to** (MMSP's routedClientType), never of\n * the catalog row: the registry carries no fast-tier capability flag, but the routing is\n * deterministic, so the answer is too. This reports what the selected client does with the\n * parameter:\n *\n * - maps it -> the protocol, and the toggle may be offered;\n * - raises UnsupportedParameterError (the Z.AI, Moonshot, DeepSeek, google-genai and\n *   embedding clients, and anthropic-official on Bedrock or for the generations that reject\n *   the `speed` parameter) -> `undefined`;\n * - forwards it to an MMSP server, whose upstream client decides what it does with it (mmsp)\n *   -> `undefined`: the harness cannot say on which protocol, or whether at all;\n * - routes nowhere (AutoLLMClient throws for an id it cannot place) -> `undefined` as well,\n *   since a model that cannot run has no fast tier either.\n *\n * A rule rather than a per-model list on purpose: catalog rows added later inherit the right\n * answer without anyone remembering to update a table.\n *\n * `\"anthropic\"` is reported for every Claude the client serves, including ids outside the\n * research preview's Opus allowlist: Anthropic answers those with a 429 at request time, which\n * is something to warn about before enabling, not grounds to hide the setting.\n *\n * Two runtime inputs stay invisible to a pure function of the config and can still flip the\n * answer: the server's `CLIENT_TYPE` env var names the client for an entry that pins none, and\n * `ANTHROPIC_BASE_URL` supplies the base URL when the entry leaves it blank (so a `bedrock://`\n * there sends Claude to Bedrock, which has no fast tier). Third-party OpenAI-compatible\n * endpoints are a third: they accept `service_tier` and may quietly serve the standard tier.\n * That residue is why llm/generative-model.ts still handles the rejection at runtime.\n */\nexport function fastModeProtocol(\n  modelId: string,\n  clientType?: string,\n  baseUrl?: string,\n): FastModeProtocol | undefined {\n  // A blank pin is none, which routes by the id.\n  const routed = routedClientType(modelId, clientType?.trim() ? clientType : null);\n  // Bedrock has no fast tier, and these Claude generations reject the `speed` parameter (MMSP\n  // 0.5.2's NO_FAST_MODE); both tests run against what the client was constructed with, as the\n  // client's own do.\n  if (\n    routed === \"anthropic-official\" &&\n    (baseUrl?.startsWith(\"bedrock://\") ||\n      [\"4-6\", \"sonnet-5-5\", \"haiku-5-5\", \"fable-5-1\"].some((generation) =>\n        modelId.includes(generation),\n      ))\n  ) {\n    return undefined;\n  }\n  return routed === null ? undefined : CLIENT_PROTOCOLS[routed]?.fastMode;\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AASA,SAAS,wBAAwB;AAIjC,SAAS,oBAAAA,yBAAwB;AAkB1B,IAAM,mBAET;AAAA,EACF,mBAAmB,EAAE,MAAM,cAAc,UAAU,SAAS;AAAA,EAC5D,sBAAsB,EAAE,MAAM,gBAAgB,UAAU,YAAY;AAAA,EACpE,mBAAmB,EAAE,MAAM,wBAAwB,UAAU,SAAS;AAAA,EACtE,gBAAgB,EAAE,MAAM,oBAAoB;AAAA,EAC5C,qBAAqB,EAAE,MAAM,oBAAoB;AAAA,EACjD,qBAAqB,EAAE,MAAM,aAAa;AAAA,EAC1C,oBAAoB,EAAE,MAAM,cAAc,UAAU,SAAS;AAAA,EAC7D,oBAAoB,EAAE,MAAM,cAAc,UAAU,SAAS;AAAA,EAC7D,eAAe,EAAE,MAAM,qBAAqB,UAAU,SAAS;AAAA,EAC/D,4BAA4B,EAAE,MAAM,qBAAqB,UAAU,SAAS;AAAA,EAC5E,oBAAoB,EAAE,MAAM,cAAc;AAAA,EAC1C,gBAAgB,EAAE,MAAM,gBAAgB,UAAU,YAAY;AAAA,EAC9D,gBAAgB,EAAE,MAAM,iBAAiB;AAAA;AAAA;AAAA,EAGzC,MAAM,EAAE,MAAM,UAAU;AAC1B;AAiCO,SAAS,iBACd,SACA,YACA,SAC8B;AAE9B,QAAM,SAAS,iBAAiB,SAAS,YAAY,KAAK,IAAI,aAAa,IAAI;AAI/E,MACE,WAAW,yBACV,SAAS,WAAW,YAAY,KAC/B,CAAC,OAAO,cAAc,aAAa,WAAW,EAAE;AAAA,IAAK,CAAC,eACpD,QAAQ,SAAS,UAAU;AAAA,EAC7B,IACF;AACA,WAAO;AAAA,EACT;AACA,SAAO,WAAW,OAAO,SAAY,iBAAiB,MAAM,GAAG;AACjE;","names":["routedClientType"]}