@plurnk/plurnk-providers 1.13.0 → 1.14.1

package/src/sdkModels.test.ts

@@ -37,21 +37,21 @@ test("{§model-catalog-readiness}: readiness uses the construction credential an
     });
     assert.doesNotMatch(JSON.stringify(providerReadiness("google", { GEMINI_API_KEY: "secret-value" })), /secret-value/);
 
-    assert.deepEqual(providerReadiness("cloudflare", { CLOUDFLARE_API_KEY: "key" }), {
+    assert.deepEqual(providerReadiness("cloudflare-workers-ai", { CLOUDFLARE_API_KEY: "key" }), {
         ready: false,
         causes: [{
             kind: "configuration",
             alternatives: [["CLOUDFLARE_ACCOUNT_ID"]],
         }],
     });
-    assert.deepEqual(providerReadiness("cloudflare", {
+    assert.deepEqual(providerReadiness("cloudflare-workers-ai", {
         CLOUDFLARE_ACCOUNT_ID: "account",
         CLOUDFLARE_API_KEY: "key",
     }), { ready: true, causes: [] });
 });
 
 test("{§model-catalog-readiness}: Bedrock reports its actual alternative authentication sets and region requirement", () => {
-    assert.deepEqual(providerReadiness("bedrock", {}), {
+    assert.deepEqual(providerReadiness("amazon-bedrock", {}), {
         ready: false,
         causes: [{
             kind: "credential",
@@ -63,11 +63,11 @@ test("{§model-catalog-readiness}: Bedrock reports its actual alternative authen
             ],
         }],
     });
-    assert.deepEqual(providerReadiness("bedrock", {
+    assert.deepEqual(providerReadiness("amazon-bedrock", {
         AWS_BEARER_TOKEN_BEDROCK: "bearer",
         AWS_REGION: "us-east-1",
     }), { ready: true, causes: [] });
-    assert.deepEqual(providerReadiness("bedrock", {
+    assert.deepEqual(providerReadiness("amazon-bedrock", {
         AWS_ACCESS_KEY_ID: "access",
         AWS_SECRET_ACCESS_KEY: "secret",
     }), {
@@ -129,14 +129,25 @@ test("{§provider-reasoning-policy} createSdkModel represents a fixed OpenRouter
         OPENROUTER_HTTP_REFERER: "https://github.com/plurnk/plurnk-service",
         OPENROUTER_APP_TITLE: "Plurnk",
     };
-    const settings = (reasoning?: "off" | "adaptive" | "low" | "medium" | "high"): unknown =>
-        (createSdkModel("openrouter", "z-ai/glm-5.3-flash", env, undefined, reasoning)?.languageModel as { settings?: { reasoning?: unknown } } | undefined)?.settings?.reasoning;
+    const settings = (reasoning?: "off" | "adaptive" | "low" | "medium" | "high", budget: number | null = null): unknown =>
+        (createSdkModel("openrouter", "z-ai/glm-5.3-flash", env, undefined, reasoning, budget)?.languageModel as { settings?: { reasoning?: unknown } } | undefined)?.settings?.reasoning;
     assert.deepEqual(settings("low"), { effort: "low" });
     assert.deepEqual(settings("medium"), { effort: "medium" });
     assert.deepEqual(settings("high"), { effort: "high" });
     assert.deepEqual(settings("off"), { effort: "none" });
-    assert.equal(settings("adaptive"), undefined);
-    assert.equal(settings(), undefined);
+    // #457 — adaptive on a reasoning-capable route declares the toggle on.
+    assert.deepEqual(settings("adaptive"), { enabled: true });
+    assert.equal(settings(), undefined, "no declared policy emits nothing");
+    // A resolved budget is the max_tokens form of the same dial; a fixed policy keeps effort.
+    assert.deepEqual(settings("adaptive", 2048), { max_tokens: 2048 });
+    assert.equal(settings(undefined, 2048), undefined, "a budget rides a policy, never replaces one");
+    assert.deepEqual(settings("low", 2048), { effort: "low" });
+    assert.deepEqual(settings("off", 2048), { effort: "none" });
+    // The budget_tokens-only motivating route (#466) and a non-reasoning route.
+    const qwen = (createSdkModel("openrouter", "qwen/qwen3.8-flash", env, undefined, "adaptive", null)?.languageModel as { settings?: { reasoning?: unknown } } | undefined)?.settings?.reasoning;
+    assert.deepEqual(qwen, { enabled: true });
+    const prose = (createSdkModel("openrouter", "tencent/hy-mt2-30b-a3b", env, undefined, "adaptive", null)?.languageModel as { settings?: { reasoning?: unknown } } | undefined)?.settings?.reasoning;
+    assert.equal(prose, undefined, "a non-reasoning route stays untouched");
 });
 
 test("{§openrouter-app-attribution} attribution rejects malformed URLs and the retired title name", () => {
@@ -183,7 +194,7 @@ test("createSdkModel attaches DeepInfra's documented response-cost normalizer",
 });
 
 test("{§provider-fact-authority} catalog credential names are law without any package or operator alias", () => {
-    const sdk = createSdkModel("cloudflare", "@cf/google/gemma-4-26b-a4b-it", {
+    const sdk = createSdkModel("cloudflare-workers-ai", "@cf/google/gemma-4-26b-a4b-it", {
         CLOUDFLARE_ACCOUNT_ID: "account",
         CLOUDFLARE_API_KEY: "key",
     });
@@ -196,10 +207,10 @@ test("{§provider-fact-authority} catalog credential names are law without any p
 });
 
 test("{§provider-fact-authority} an operator credential override holds one exact name", () => {
-    const sdk = createSdkModel("cloudflare", "@cf/google/gemma-4-26b-a4b-it", {
+    const sdk = createSdkModel("cloudflare-workers-ai", "@cf/google/gemma-4-26b-a4b-it", {
         CLOUDFLARE_ACCOUNT_ID: "account",
         MY_ORG_CLOUDFLARE_KEY: "key",
-        PLURNK_PROVIDERS_PROVIDER_CLOUDFLARE_API_KEY_ENV: "MY_ORG_CLOUDFLARE_KEY",
+        PLURNK_PROVIDERS_PROVIDER_CLOUDFLARE_WORKERS_AI_API_KEY_ENV: "MY_ORG_CLOUDFLARE_KEY",
     });
     assert.deepEqual(sdk?.compatible, {
         url: "https://api.cloudflare.com/client/v4/accounts/account/ai/v1/chat/completions",
@@ -209,10 +220,10 @@ test("{§provider-fact-authority} an operator credential override holds one exac
 
 test("{§provider-fact-authority} ordered credential fallbacks are rejected at construction", () => {
     assert.throws(
-        () => createSdkModel("cloudflare", "@cf/google/gemma-4-26b-a4b-it", {
+        () => createSdkModel("cloudflare-workers-ai", "@cf/google/gemma-4-26b-a4b-it", {
             CLOUDFLARE_ACCOUNT_ID: "account",
             CLOUDFLARE_API_TOKEN: "token",
-            PLURNK_PROVIDERS_PROVIDER_CLOUDFLARE_API_KEY_ENV: "CLOUDFLARE_API_TOKEN,CLOUDFLARE_API_KEY",
+            PLURNK_PROVIDERS_PROVIDER_CLOUDFLARE_WORKERS_AI_API_KEY_ENV: "CLOUDFLARE_API_TOKEN,CLOUDFLARE_API_KEY",
         }),
         /one exact name, never an ordered fallback/,
     );
@@ -239,7 +250,7 @@ test("catalog routes own their documented cache-affinity request projection", ()
         { target: "header", name: "x-session-id" },
     );
     assert.deepEqual(
-        createSdkModel("fireworks", "accounts/fireworks/models/test", { FIREWORKS_API_KEY: "key" })?.cacheAffinity,
+        createSdkModel("fireworks-ai", "accounts/fireworks/models/test", { FIREWORKS_API_KEY: "key" })?.cacheAffinity,
         { target: "body", name: "prompt_cache_key" },
     );
 });
@@ -292,19 +303,19 @@ test("createSdkModel fails clearly for a declared but unsupported SDK package",
 });
 
 test("{§provider-embedding-resolution} compatible catalog and declared routes share the standard embedding adapter", () => {
-    const cloudflare = createEmbeddingModel("cloudflare", "@cf/qwen/qwen3-embedding-0.6b", {
+    const cloudflare = createEmbeddingModel("cloudflare-workers-ai", "@cf/qwen/qwen3-embedding-0.6b", {
         CLOUDFLARE_ACCOUNT_ID: "account",
         CLOUDFLARE_API_KEY: "key",
     });
     assert.equal(cloudflare?.providerId, "cloudflare-workers-ai");
-    assert.equal(cloudflare?.embeddingModel.provider, "cloudflare.embedding");
+    assert.equal(cloudflare?.embeddingModel.provider, "cloudflare-workers-ai.embedding");
     assert.equal(cloudflare?.embeddingModel.modelId, "@cf/qwen/qwen3-embedding-0.6b");
 
-    const fireworks = createEmbeddingModel("fireworks", "fireworks/qwen3-embedding-8b", {
+    const fireworks = createEmbeddingModel("fireworks-ai", "fireworks/qwen3-embedding-8b", {
         FIREWORKS_API_KEY: "key",
     });
     assert.equal(fireworks?.providerId, "fireworks-ai");
-    assert.equal(fireworks?.embeddingModel.provider, "fireworks.embedding");
+    assert.equal(fireworks?.embeddingModel.provider, "fireworks-ai.embedding");
 
     const local = createEmbeddingModel("local-embed", "Qwen/Qwen3-Embedding-0.6B", {
         PLURNK_PROVIDERS_PROVIDER_LOCAL_EMBED_NPM: "@ai-sdk/openai-compatible",

package/src/sdkModels.ts

@@ -10,8 +10,9 @@ import { createOpenAICompatible } from "@ai-sdk/openai-compatible";
 import { createTogetherAI } from "@ai-sdk/togetherai";
 import { createXai } from "@ai-sdk/xai";
 import type { EmbeddingModelV4, EmbeddingModelV4Result } from "@ai-sdk/provider";
-import { createOpenRouter } from "@openrouter/ai-sdk-provider";
+import { createOpenRouter, type OpenRouterChatSettings } from "@openrouter/ai-sdk-provider";
 import {
+    resolveModel,
     isProviderCredentialName,
     lookup,
     lookupProvider,
@@ -310,14 +311,31 @@ const requireApiKey = (
 
 // {§provider-reasoning-policy} — the OpenRouter SDK builds its request from the model
 // settings alone (neither the generic `reasoning` call setting nor provider options reach
-// its body), so a fixed policy and `off` are represented there; `adaptive` leaves the
-// route's provider default.
+// its body), so a fixed policy and `off` are represented there. `adaptive` on a
+// reasoning-capable route declares the toggle on (`enabled` — #457: unconfigured
+// toggle routes ran reasoning-off and strike-spiralled; `off` stays the explicit
+// opt-out), and a resolved reasoning budget names the wire's `max_tokens` form —
+// the only dial a budget_tokens-only route understands; a fixed policy keeps the
+// `effort` form. `enabled` alone is documented OpenRouter wire; the SDK's settings
+// type under-models it (requires effort | max_tokens), hence the call-site cast.
 const openRouterReasoningSettings = (
     reasoning: ReasoningPolicy | undefined,
-): { reasoning?: { effort: "none" | "low" | "medium" | "high" } } =>
-    reasoning === undefined || reasoning === "adaptive"
-        ? {}
-        : { reasoning: { effort: reasoning === "off" ? "none" : reasoning } };
+    reasoningBudget: number | null,
+    reasoningCapable: boolean,
+): { reasoning?: { effort: "none" | "low" | "medium" | "high" } | { max_tokens: number } | { enabled: true } } => {
+    if (reasoning === undefined) return {};
+    if (reasoning === "adaptive") {
+        if (reasoningBudget !== null) return { reasoning: { max_tokens: reasoningBudget } };
+        return reasoningCapable ? { reasoning: { enabled: true } } : {};
+    }
+    // OpenRouter documents effort none/low/medium/high; xhigh/max reach this
+    // seam only if a catalog row declares them — refuse with the gap named
+    // rather than ship an unverified wire value (#474 widens on evidence).
+    if (reasoning === "xhigh" || reasoning === "max") {
+        throw new TypeError(`openrouter reasoning effort '${reasoning}' is not a documented wire value`);
+    }
+    return { reasoning: { effort: reasoning === "off" ? "none" : reasoning } };
+};
 
 const resolveSdkProvider = (
     provider: string,
@@ -339,6 +357,7 @@ export const createSdkModel = (
     env: NodeJS.ProcessEnv,
     baseUrlOverride?: string,
     reasoning?: ReasoningPolicy,
+    reasoningBudget: number | null = null,
 ): SdkModel | null => {
     const resolved = resolveSdkProvider(provider, env, baseUrlOverride);
     if (resolved === null) return null;
@@ -430,7 +449,11 @@ export const createSdkModel = (
                     apiKey: requireApiKey(provider, env, catalog),
                     baseURL: url,
                     headers: openRouterHeaders(provider, env, catalog),
-                }).languageModel(model, openRouterReasoningSettings(reasoning)),
+                }).languageModel(model, openRouterReasoningSettings(
+                    reasoning,
+                    reasoningBudget,
+                    resolveModel(provider, model)?.info.reasoning === true,
+                ) as OpenRouterChatSettings),
                 ...(catalog.id === "openrouter"
                     ? { cacheAffinity: { target: "header" as const, name: "x-session-id" } }
                     : {}),