@neon/ai-sdk-provider 0.6.0 → 0.7.1

package/README.md

@@ -88,7 +88,7 @@ for await (const part of result.fullStream) {
 ## Limitations
 
 - `generateImage()` and embeddings (`embed` / `embedMany`) are not offered by the gateway and throw `NoSuchModelError`.
-- `gpt-oss-*` models return a non-standard ("harmony") response shape on the unified endpoint and are not fully supported.
+- `gpt-oss-*` models return a non-standard ("harmony") response shape on the unified endpoint (`message.content` as an array of reasoning/text parts instead of a string). The provider normalizes this to the OpenAI Chat Completions contract (string `content` + `reasoning_content`) so `generateText`/`streamText` work and reasoning is surfaced. See neondatabase/neon-pkgs#308.
 - OpenAI Responses multi-turn tool flows (`generateText` + `stepCountIs`) can return 502 from the gateway; tool calling is covered on Anthropic/Google/Meta in e2e.
 
 ## End-to-end tests

package/dist/lib/neon-anthropic-language-model.d.ts

@@ -5,7 +5,7 @@ import { LanguageModelV3, LanguageModelV3CallOptions, LanguageModelV3GenerateRes
 
 /**
  * Language model for Anthropic models served via the Neon AI Gateway's native
- * Messages API (`/ai-gateway/anthropic/v1/messages`). Unlocks streaming
+ * Messages API (`/anthropic/v1/messages`). Unlocks streaming
  * structured output and native reasoning for Claude.
  *
  * The shared Anthropic model defaults to fine-grained tool-input streaming

package/dist/lib/neon-anthropic-language-model.js

@@ -2,7 +2,7 @@ import { AnthropicMessagesLanguageModel } from "@ai-sdk/anthropic/internal";
 //#region src/lib/neon-anthropic-language-model.ts
 /**
 * Language model for Anthropic models served via the Neon AI Gateway's native
-* Messages API (`/ai-gateway/anthropic/v1/messages`). Unlocks streaming
+* Messages API (`/anthropic/v1/messages`). Unlocks streaming
 * structured output and native reasoning for Claude.
 *
 * The shared Anthropic model defaults to fine-grained tool-input streaming

package/dist/lib/neon-anthropic-language-model.js.map

@@ -1 +1 @@
-{"version":3,"file":"neon-anthropic-language-model.js","names":[],"sources":["../../src/lib/neon-anthropic-language-model.ts"],"sourcesContent":["import { AnthropicMessagesLanguageModel } from \"@ai-sdk/anthropic/internal\";\nimport type {\n\tLanguageModelV3,\n\tLanguageModelV3CallOptions,\n\tLanguageModelV3GenerateResult,\n\tLanguageModelV3StreamResult,\n} from \"@ai-sdk/provider\";\n\n/**\n * Language model for Anthropic models served via the Neon AI Gateway's native\n * Messages API (`/ai-gateway/anthropic/v1/messages`). Unlocks streaming\n * structured output and native reasoning for Claude.\n *\n * The shared Anthropic model defaults to fine-grained tool-input streaming\n * (`eager_input_streaming: true`) on streaming tool calls, which the gateway\n * rejects (`Extra inputs are not permitted`). We disable it via the model's own\n * `toolStreaming` option so streaming tool calls work. Users can still override.\n */\nexport class NeonAnthropicLanguageModel\n\textends AnthropicMessagesLanguageModel\n\timplements LanguageModelV3\n{\n\tprivate withGatewayCompat(\n\t\toptions: LanguageModelV3CallOptions,\n\t): LanguageModelV3CallOptions {\n\t\tconst anthropic = options.providerOptions?.anthropic;\n\t\t// Respect an explicit user setting.\n\t\tif (anthropic != null && \"toolStreaming\" in anthropic) {\n\t\t\treturn options;\n\t\t}\n\t\treturn {\n\t\t\t...options,\n\t\t\tproviderOptions: {\n\t\t\t\t...options.providerOptions,\n\t\t\t\tanthropic: { ...anthropic, toolStreaming: false },\n\t\t\t},\n\t\t};\n\t}\n\n\toverride doGenerate(\n\t\toptions: LanguageModelV3CallOptions,\n\t): Promise<LanguageModelV3GenerateResult> {\n\t\treturn super.doGenerate(this.withGatewayCompat(options));\n\t}\n\n\toverride doStream(\n\t\toptions: LanguageModelV3CallOptions,\n\t): Promise<LanguageModelV3StreamResult> {\n\t\treturn super.doStream(this.withGatewayCompat(options));\n\t}\n}\n"],"mappings":";;;;;;;;;;;;AAkBA,IAAa,6BAAb,cACS,+BAET;CACC,kBACC,SAC6B;EAC7B,MAAM,YAAY,QAAQ,iBAAiB;EAE3C,IAAI,aAAa,QAAQ,mBAAmB,WAC3C,OAAO;EAER,OAAO;GACN,GAAG;GACH,iBAAiB;IAChB,GAAG,QAAQ;IACX,WAAW;KAAE,GAAG;KAAW,eAAe;IAAM;GACjD;EACD;CACD;CAEA,WACC,SACyC;EACzC,OAAO,MAAM,WAAW,KAAK,kBAAkB,OAAO,CAAC;CACxD;CAEA,SACC,SACuC;EACvC,OAAO,MAAM,SAAS,KAAK,kBAAkB,OAAO,CAAC;CACtD;AACD"}
+{"version":3,"file":"neon-anthropic-language-model.js","names":[],"sources":["../../src/lib/neon-anthropic-language-model.ts"],"sourcesContent":["import { AnthropicMessagesLanguageModel } from \"@ai-sdk/anthropic/internal\";\nimport type {\n\tLanguageModelV3,\n\tLanguageModelV3CallOptions,\n\tLanguageModelV3GenerateResult,\n\tLanguageModelV3StreamResult,\n} from \"@ai-sdk/provider\";\n\n/**\n * Language model for Anthropic models served via the Neon AI Gateway's native\n * Messages API (`/anthropic/v1/messages`). Unlocks streaming\n * structured output and native reasoning for Claude.\n *\n * The shared Anthropic model defaults to fine-grained tool-input streaming\n * (`eager_input_streaming: true`) on streaming tool calls, which the gateway\n * rejects (`Extra inputs are not permitted`). We disable it via the model's own\n * `toolStreaming` option so streaming tool calls work. Users can still override.\n */\nexport class NeonAnthropicLanguageModel\n\textends AnthropicMessagesLanguageModel\n\timplements LanguageModelV3\n{\n\tprivate withGatewayCompat(\n\t\toptions: LanguageModelV3CallOptions,\n\t): LanguageModelV3CallOptions {\n\t\tconst anthropic = options.providerOptions?.anthropic;\n\t\t// Respect an explicit user setting.\n\t\tif (anthropic != null && \"toolStreaming\" in anthropic) {\n\t\t\treturn options;\n\t\t}\n\t\treturn {\n\t\t\t...options,\n\t\t\tproviderOptions: {\n\t\t\t\t...options.providerOptions,\n\t\t\t\tanthropic: { ...anthropic, toolStreaming: false },\n\t\t\t},\n\t\t};\n\t}\n\n\toverride doGenerate(\n\t\toptions: LanguageModelV3CallOptions,\n\t): Promise<LanguageModelV3GenerateResult> {\n\t\treturn super.doGenerate(this.withGatewayCompat(options));\n\t}\n\n\toverride doStream(\n\t\toptions: LanguageModelV3CallOptions,\n\t): Promise<LanguageModelV3StreamResult> {\n\t\treturn super.doStream(this.withGatewayCompat(options));\n\t}\n}\n"],"mappings":";;;;;;;;;;;;AAkBA,IAAa,6BAAb,cACS,+BAET;CACC,kBACC,SAC6B;EAC7B,MAAM,YAAY,QAAQ,iBAAiB;EAE3C,IAAI,aAAa,QAAQ,mBAAmB,WAC3C,OAAO;EAER,OAAO;GACN,GAAG;GACH,iBAAiB;IAChB,GAAG,QAAQ;IACX,WAAW;KAAE,GAAG;KAAW,eAAe;IAAM;GACjD;EACD;CACD;CAEA,WACC,SACyC;EACzC,OAAO,MAAM,WAAW,KAAK,kBAAkB,OAAO,CAAC;CACxD;CAEA,SACC,SACuC;EACvC,OAAO,MAAM,SAAS,KAAK,kBAAkB,OAAO,CAAC;CACtD;AACD"}

package/dist/lib/neon-harmony-normalize.d.ts

@@ -0,0 +1,41 @@
+import { FetchFunction } from "@ai-sdk/provider-utils";
+
+//#region src/lib/neon-harmony-normalize.d.ts
+interface ExtractedHarmony {
+  text: string;
+  reasoning: string;
+}
+/** The result of a normalization pass, flagging whether anything was rewritten. */
+interface NormalizeResult {
+  body: unknown;
+  changed: boolean;
+}
+/**
+ * Pull the text and reasoning out of a harmony `content` array. Returns `null`
+ * when `content` is not an array (i.e. an already-compliant string response),
+ * signalling that no rewrite is needed.
+ */
+declare function extractHarmonyContent(content: unknown): ExtractedHarmony | null;
+/**
+ * Rewrite a non-streaming `chat.completion` body in place. `changed` is false
+ * when the body was already compliant (no harmony array found).
+ */
+declare function normalizeCompletionBody(body: unknown): NormalizeResult;
+/**
+ * Rewrite a streaming `chat.completion.chunk` body in place. `changed` is false
+ * when the chunk was already compliant (no harmony array found).
+ */
+declare function normalizeChunkBody(body: unknown): NormalizeResult;
+/**
+ * Wrap a fetch implementation so gpt-oss harmony responses are normalized to
+ * the OpenAI Chat Completions shape before the caller parses them. Composes
+ * with a user-supplied `fetch` (falls back to the global `fetch`).
+ *
+ * When the response is already compliant, the original response is returned
+ * untouched — a transparent pass-through that will not interfere once the
+ * gateway is fixed.
+ */
+declare function wrapFetchWithHarmonyNormalization(baseFetch?: FetchFunction): FetchFunction;
+//#endregion
+export { extractHarmonyContent, normalizeChunkBody, normalizeCompletionBody, wrapFetchWithHarmonyNormalization };
+//# sourceMappingURL=neon-harmony-normalize.d.ts.map

package/dist/lib/neon-harmony-normalize.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"neon-harmony-normalize.d.ts","names":[],"sources":["../../src/lib/neon-harmony-normalize.ts"],"mappings":";;;UAwCU,gBAAA;;EAAA,SAAA,EAAA,MAAA;AAAgB;AAgB1B;AAkDA,UA5DU,eAAA,CA4DM;EAkCA,IAAA,EAAA,OAAA;EAuGA,OAAA,EAAA,OAAA;;;;AAEA;;;iBA7LA,qBAAA,oBAEb;;;;;iBAgDa,uBAAA,iBAAwC;;;;;iBAkCxC,kBAAA,iBAAmC;;;;;;;;;;iBAuGnC,iCAAA,aACH,gBACV"}

package/dist/lib/neon-harmony-normalize.js

@@ -0,0 +1,198 @@
+//#region src/lib/neon-harmony-normalize.ts
+/**
+* Normalize the Neon AI Gateway's non-OpenAI-compliant `gpt-oss` ("harmony")
+* response shape into the OpenAI Chat Completions contract before it reaches
+* the AI SDK's schema validation.
+*
+* For `gpt-oss-*` the gateway returns `message.content` (and streaming
+* `delta.content`) as an array of Responses-style parts:
+*
+*   "content": [
+*     { "type": "reasoning", "summary": [{ "type": "summary_text", "text": "…" }] },
+*     { "type": "text", "text": "…" }
+*   ]
+*
+* The Chat Completions spec requires `content` to be a string, so any strict
+* OpenAI-compatible client (incl. the Vercel AI SDK) rejects it. We flatten the
+* `text` parts into the string `content` and hoist reasoning into
+* `reasoning_content` (which `@ai-sdk/openai-compatible` already surfaces as an
+* AI SDK reasoning part).
+*
+* The transform only rewrites when `content` is an array, and the fetch wrapper
+* returns the response untouched when nothing was rewritten. So it is a true
+* no-op / pass-through for every spec-compliant model (Llama, Gemini, Qwen, …)
+* and self-retires the moment the gateway starts returning a compliant shape —
+* it will not double-transform or otherwise interfere with the downstream fix.
+*
+* See neondatabase/neon-pkgs#308.
+*/
+function isRecord(value) {
+	return typeof value === "object" && value !== null && !Array.isArray(value);
+}
+/**
+* Pull the text and reasoning out of a harmony `content` array. Returns `null`
+* when `content` is not an array (i.e. an already-compliant string response),
+* signalling that no rewrite is needed.
+*/
+function extractHarmonyContent(content) {
+	if (!Array.isArray(content)) return null;
+	const texts = [];
+	const reasonings = [];
+	for (const part of content) {
+		if (!isRecord(part)) continue;
+		if (part.type === "text") {
+			if (typeof part.text === "string") texts.push(part.text);
+			continue;
+		}
+		if (part.type === "reasoning") {
+			if (typeof part.text === "string") reasonings.push(part.text);
+			if (Array.isArray(part.summary)) {
+				for (const entry of part.summary) if (isRecord(entry) && typeof entry.text === "string") reasonings.push(entry.text);
+			}
+			if (Array.isArray(part.content)) {
+				for (const entry of part.content) if (isRecord(entry) && typeof entry.text === "string") reasonings.push(entry.text);
+			}
+		}
+	}
+	return {
+		text: texts.join(""),
+		reasoning: reasonings.join("\n")
+	};
+}
+/**
+* Rewrite a non-streaming `chat.completion` body in place. `changed` is false
+* when the body was already compliant (no harmony array found).
+*/
+function normalizeCompletionBody(body) {
+	if (!isRecord(body) || !Array.isArray(body.choices)) return {
+		body,
+		changed: false
+	};
+	let changed = false;
+	for (const choice of body.choices) {
+		if (!isRecord(choice)) continue;
+		const message = choice.message;
+		if (!isRecord(message)) continue;
+		const extracted = extractHarmonyContent(message.content);
+		if (extracted === null) continue;
+		message.content = extracted.text;
+		changed = true;
+		if (extracted.reasoning.length > 0 && message.reasoning_content == null && message.reasoning == null) message.reasoning_content = extracted.reasoning;
+	}
+	return {
+		body,
+		changed
+	};
+}
+/**
+* Rewrite a streaming `chat.completion.chunk` body in place. `changed` is false
+* when the chunk was already compliant (no harmony array found).
+*/
+function normalizeChunkBody(body) {
+	if (!isRecord(body) || !Array.isArray(body.choices)) return {
+		body,
+		changed: false
+	};
+	let changed = false;
+	for (const choice of body.choices) {
+		if (!isRecord(choice)) continue;
+		const delta = choice.delta;
+		if (!isRecord(delta)) continue;
+		const extracted = extractHarmonyContent(delta.content);
+		if (extracted === null) continue;
+		changed = true;
+		if (extracted.text.length > 0) delta.content = extracted.text;
+		else delta.content = void 0;
+		if (extracted.reasoning.length > 0 && delta.reasoning_content == null && delta.reasoning == null) delta.reasoning_content = extracted.reasoning;
+	}
+	return {
+		body,
+		changed
+	};
+}
+/**
+* Rewrite each `data:` frame of an SSE stream. Compliant frames pass through
+* verbatim; only harmony frames are re-serialized.
+*/
+function normalizeEventStream(body) {
+	const decoder = new TextDecoder();
+	const encoder = new TextEncoder();
+	let buffer = "";
+	const rewriteLine = (line) => {
+		if (!line.startsWith("data:")) return line;
+		const payload = line.slice(5).trim();
+		if (payload === "" || payload === "[DONE]") return line;
+		let parsed;
+		try {
+			parsed = JSON.parse(payload);
+		} catch {
+			return line;
+		}
+		const { body: normalized, changed } = normalizeChunkBody(parsed);
+		return changed ? `data: ${JSON.stringify(normalized)}` : line;
+	};
+	return body.pipeThrough(new TransformStream({
+		transform(chunk, controller) {
+			buffer += decoder.decode(chunk, { stream: true });
+			const lines = buffer.split("\n");
+			buffer = lines.pop() ?? "";
+			for (const line of lines) controller.enqueue(encoder.encode(`${rewriteLine(line)}\n`));
+		},
+		flush(controller) {
+			if (buffer.length > 0) controller.enqueue(encoder.encode(rewriteLine(buffer)));
+		}
+	}));
+}
+function headersWithout(source, ...names) {
+	const copy = new Headers(source);
+	for (const name of names) copy.delete(name);
+	return copy;
+}
+/**
+* Wrap a fetch implementation so gpt-oss harmony responses are normalized to
+* the OpenAI Chat Completions shape before the caller parses them. Composes
+* with a user-supplied `fetch` (falls back to the global `fetch`).
+*
+* When the response is already compliant, the original response is returned
+* untouched — a transparent pass-through that will not interfere once the
+* gateway is fixed.
+*/
+function wrapFetchWithHarmonyNormalization(baseFetch) {
+	const doFetch = baseFetch ?? ((input, init) => globalThis.fetch(input, init));
+	return async (input, init) => {
+		const response = await doFetch(input, init);
+		if (!response.ok || response.body === null) return response;
+		const contentType = response.headers.get("content-type") ?? "";
+		if (contentType.includes("text/event-stream")) return new Response(normalizeEventStream(response.body), {
+			status: response.status,
+			statusText: response.statusText,
+			headers: headersWithout(response.headers, "content-length", "content-encoding")
+		});
+		if (contentType.includes("application/json")) {
+			let text;
+			try {
+				text = await response.clone().text();
+			} catch {
+				return response;
+			}
+			let parsed;
+			try {
+				parsed = JSON.parse(text);
+			} catch {
+				return response;
+			}
+			const { body: normalized, changed } = normalizeCompletionBody(parsed);
+			if (!changed) return response;
+			return new Response(JSON.stringify(normalized), {
+				status: response.status,
+				statusText: response.statusText,
+				headers: headersWithout(response.headers, "content-length", "content-encoding")
+			});
+		}
+		return response;
+	};
+}
+//#endregion
+export { extractHarmonyContent, normalizeChunkBody, normalizeCompletionBody, wrapFetchWithHarmonyNormalization };
+
+//# sourceMappingURL=neon-harmony-normalize.js.map

package/dist/lib/neon-harmony-normalize.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"neon-harmony-normalize.js","names":[],"sources":["../../src/lib/neon-harmony-normalize.ts"],"sourcesContent":["import type { FetchFunction } from \"@ai-sdk/provider-utils\";\n\n// TODO(#308): Temporary workaround. Remove this whole module (and its wiring in\n// provider.ts) once the Neon AI Gateway returns Chat Completions-compliant\n// gpt-oss responses (string `content` + `reasoning_content`). The wrapper is a\n// transparent pass-through for any already-compliant response, so it is safe to\n// leave in place until the gateway fix ships — but it should be deleted then.\n\n/**\n * Normalize the Neon AI Gateway's non-OpenAI-compliant `gpt-oss` (\"harmony\")\n * response shape into the OpenAI Chat Completions contract before it reaches\n * the AI SDK's schema validation.\n *\n * For `gpt-oss-*` the gateway returns `message.content` (and streaming\n * `delta.content`) as an array of Responses-style parts:\n *\n *   \"content\": [\n *     { \"type\": \"reasoning\", \"summary\": [{ \"type\": \"summary_text\", \"text\": \"…\" }] },\n *     { \"type\": \"text\", \"text\": \"…\" }\n *   ]\n *\n * The Chat Completions spec requires `content` to be a string, so any strict\n * OpenAI-compatible client (incl. the Vercel AI SDK) rejects it. We flatten the\n * `text` parts into the string `content` and hoist reasoning into\n * `reasoning_content` (which `@ai-sdk/openai-compatible` already surfaces as an\n * AI SDK reasoning part).\n *\n * The transform only rewrites when `content` is an array, and the fetch wrapper\n * returns the response untouched when nothing was rewritten. So it is a true\n * no-op / pass-through for every spec-compliant model (Llama, Gemini, Qwen, …)\n * and self-retires the moment the gateway starts returning a compliant shape —\n * it will not double-transform or otherwise interfere with the downstream fix.\n *\n * See neondatabase/neon-pkgs#308.\n */\n\nfunction isRecord(value: unknown): value is Record<string, unknown> {\n\treturn typeof value === \"object\" && value !== null && !Array.isArray(value);\n}\n\ninterface ExtractedHarmony {\n\ttext: string;\n\treasoning: string;\n}\n\n/** The result of a normalization pass, flagging whether anything was rewritten. */\ninterface NormalizeResult {\n\tbody: unknown;\n\tchanged: boolean;\n}\n\n/**\n * Pull the text and reasoning out of a harmony `content` array. Returns `null`\n * when `content` is not an array (i.e. an already-compliant string response),\n * signalling that no rewrite is needed.\n */\nexport function extractHarmonyContent(\n\tcontent: unknown,\n): ExtractedHarmony | null {\n\tif (!Array.isArray(content)) {\n\t\treturn null;\n\t}\n\n\tconst texts: string[] = [];\n\tconst reasonings: string[] = [];\n\n\tfor (const part of content) {\n\t\tif (!isRecord(part)) {\n\t\t\tcontinue;\n\t\t}\n\t\tif (part.type === \"text\") {\n\t\t\tif (typeof part.text === \"string\") {\n\t\t\t\ttexts.push(part.text);\n\t\t\t}\n\t\t\tcontinue;\n\t\t}\n\t\tif (part.type === \"reasoning\") {\n\t\t\t// Reasoning items can carry the raw CoT in `text`, a `summary[]` of\n\t\t\t// `summary_text` parts, and/or a `content[]` of `reasoning_text` parts.\n\t\t\tif (typeof part.text === \"string\") {\n\t\t\t\treasonings.push(part.text);\n\t\t\t}\n\t\t\tif (Array.isArray(part.summary)) {\n\t\t\t\tfor (const entry of part.summary) {\n\t\t\t\t\tif (isRecord(entry) && typeof entry.text === \"string\") {\n\t\t\t\t\t\treasonings.push(entry.text);\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t\tif (Array.isArray(part.content)) {\n\t\t\t\tfor (const entry of part.content) {\n\t\t\t\t\tif (isRecord(entry) && typeof entry.text === \"string\") {\n\t\t\t\t\t\treasonings.push(entry.text);\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\treturn { text: texts.join(\"\"), reasoning: reasonings.join(\"\\n\") };\n}\n\n/**\n * Rewrite a non-streaming `chat.completion` body in place. `changed` is false\n * when the body was already compliant (no harmony array found).\n */\nexport function normalizeCompletionBody(body: unknown): NormalizeResult {\n\tif (!isRecord(body) || !Array.isArray(body.choices)) {\n\t\treturn { body, changed: false };\n\t}\n\tlet changed = false;\n\tfor (const choice of body.choices) {\n\t\tif (!isRecord(choice)) {\n\t\t\tcontinue;\n\t\t}\n\t\tconst message = choice.message;\n\t\tif (!isRecord(message)) {\n\t\t\tcontinue;\n\t\t}\n\t\tconst extracted = extractHarmonyContent(message.content);\n\t\tif (extracted === null) {\n\t\t\tcontinue;\n\t\t}\n\t\tmessage.content = extracted.text;\n\t\tchanged = true;\n\t\tif (\n\t\t\textracted.reasoning.length > 0 &&\n\t\t\tmessage.reasoning_content == null &&\n\t\t\tmessage.reasoning == null\n\t\t) {\n\t\t\tmessage.reasoning_content = extracted.reasoning;\n\t\t}\n\t}\n\treturn { body, changed };\n}\n\n/**\n * Rewrite a streaming `chat.completion.chunk` body in place. `changed` is false\n * when the chunk was already compliant (no harmony array found).\n */\nexport function normalizeChunkBody(body: unknown): NormalizeResult {\n\tif (!isRecord(body) || !Array.isArray(body.choices)) {\n\t\treturn { body, changed: false };\n\t}\n\tlet changed = false;\n\tfor (const choice of body.choices) {\n\t\tif (!isRecord(choice)) {\n\t\t\tcontinue;\n\t\t}\n\t\tconst delta = choice.delta;\n\t\tif (!isRecord(delta)) {\n\t\t\tcontinue;\n\t\t}\n\t\tconst extracted = extractHarmonyContent(delta.content);\n\t\tif (extracted === null) {\n\t\t\tcontinue;\n\t\t}\n\t\tchanged = true;\n\t\t// Drop empty text so reasoning-only chunks don't emit spurious text deltas.\n\t\tif (extracted.text.length > 0) {\n\t\t\tdelta.content = extracted.text;\n\t\t} else {\n\t\t\tdelta.content = undefined;\n\t\t}\n\t\tif (\n\t\t\textracted.reasoning.length > 0 &&\n\t\t\tdelta.reasoning_content == null &&\n\t\t\tdelta.reasoning == null\n\t\t) {\n\t\t\tdelta.reasoning_content = extracted.reasoning;\n\t\t}\n\t}\n\treturn { body, changed };\n}\n\n/**\n * Rewrite each `data:` frame of an SSE stream. Compliant frames pass through\n * verbatim; only harmony frames are re-serialized.\n */\nfunction normalizeEventStream(\n\tbody: ReadableStream<Uint8Array>,\n): ReadableStream<Uint8Array> {\n\tconst decoder = new TextDecoder();\n\tconst encoder = new TextEncoder();\n\tlet buffer = \"\";\n\n\tconst rewriteLine = (line: string): string => {\n\t\tif (!line.startsWith(\"data:\")) {\n\t\t\treturn line;\n\t\t}\n\t\tconst payload = line.slice(\"data:\".length).trim();\n\t\tif (payload === \"\" || payload === \"[DONE]\") {\n\t\t\treturn line;\n\t\t}\n\t\tlet parsed: unknown;\n\t\ttry {\n\t\t\tparsed = JSON.parse(payload);\n\t\t} catch {\n\t\t\treturn line;\n\t\t}\n\t\tconst { body: normalized, changed } = normalizeChunkBody(parsed);\n\t\t// Pass compliant frames through unchanged so we don't reshape a fixed gateway.\n\t\treturn changed ? `data: ${JSON.stringify(normalized)}` : line;\n\t};\n\n\treturn body.pipeThrough(\n\t\tnew TransformStream<Uint8Array, Uint8Array>({\n\t\t\ttransform(chunk, controller) {\n\t\t\t\tbuffer += decoder.decode(chunk, { stream: true });\n\t\t\t\tconst lines = buffer.split(\"\\n\");\n\t\t\t\tbuffer = lines.pop() ?? \"\";\n\t\t\t\tfor (const line of lines) {\n\t\t\t\t\tcontroller.enqueue(\n\t\t\t\t\t\tencoder.encode(`${rewriteLine(line)}\\n`),\n\t\t\t\t\t);\n\t\t\t\t}\n\t\t\t},\n\t\t\tflush(controller) {\n\t\t\t\tif (buffer.length > 0) {\n\t\t\t\t\tcontroller.enqueue(encoder.encode(rewriteLine(buffer)));\n\t\t\t\t}\n\t\t\t},\n\t\t}),\n\t);\n}\n\nfunction headersWithout(source: Headers, ...names: string[]): Headers {\n\tconst copy = new Headers(source);\n\tfor (const name of names) {\n\t\tcopy.delete(name);\n\t}\n\treturn copy;\n}\n\n/**\n * Wrap a fetch implementation so gpt-oss harmony responses are normalized to\n * the OpenAI Chat Completions shape before the caller parses them. Composes\n * with a user-supplied `fetch` (falls back to the global `fetch`).\n *\n * When the response is already compliant, the original response is returned\n * untouched — a transparent pass-through that will not interfere once the\n * gateway is fixed.\n */\nexport function wrapFetchWithHarmonyNormalization(\n\tbaseFetch?: FetchFunction,\n): FetchFunction {\n\tconst doFetch: FetchFunction =\n\t\tbaseFetch ?? ((input, init) => globalThis.fetch(input, init));\n\n\treturn async (input, init) => {\n\t\tconst response = await doFetch(input, init);\n\t\tif (!response.ok || response.body === null) {\n\t\t\treturn response;\n\t\t}\n\t\tconst contentType = response.headers.get(\"content-type\") ?? \"\";\n\n\t\tif (contentType.includes(\"text/event-stream\")) {\n\t\t\t// Streaming is inspected frame-by-frame; compliant frames pass through\n\t\t\t// verbatim (see normalizeEventStream).\n\t\t\treturn new Response(normalizeEventStream(response.body), {\n\t\t\t\tstatus: response.status,\n\t\t\t\tstatusText: response.statusText,\n\t\t\t\t// content-length/encoding no longer describe the (possibly) rewritten body.\n\t\t\t\theaders: headersWithout(\n\t\t\t\t\tresponse.headers,\n\t\t\t\t\t\"content-length\",\n\t\t\t\t\t\"content-encoding\",\n\t\t\t\t),\n\t\t\t});\n\t\t}\n\n\t\tif (contentType.includes(\"application/json\")) {\n\t\t\t// Inspect a clone so the original body stays intact for the pass-through case.\n\t\t\tlet text: string;\n\t\t\ttry {\n\t\t\t\ttext = await response.clone().text();\n\t\t\t} catch {\n\t\t\t\treturn response;\n\t\t\t}\n\t\t\tlet parsed: unknown;\n\t\t\ttry {\n\t\t\t\tparsed = JSON.parse(text);\n\t\t\t} catch {\n\t\t\t\treturn response;\n\t\t\t}\n\t\t\tconst { body: normalized, changed } =\n\t\t\t\tnormalizeCompletionBody(parsed);\n\t\t\tif (!changed) {\n\t\t\t\t// Already compliant: return the original response untouched.\n\t\t\t\treturn response;\n\t\t\t}\n\t\t\treturn new Response(JSON.stringify(normalized), {\n\t\t\t\tstatus: response.status,\n\t\t\t\tstatusText: response.statusText,\n\t\t\t\theaders: headersWithout(\n\t\t\t\t\tresponse.headers,\n\t\t\t\t\t\"content-length\",\n\t\t\t\t\t\"content-encoding\",\n\t\t\t\t),\n\t\t\t});\n\t\t}\n\n\t\treturn response;\n\t};\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;AAoCA,SAAS,SAAS,OAAkD;CACnE,OAAO,OAAO,UAAU,YAAY,UAAU,QAAQ,CAAC,MAAM,QAAQ,KAAK;AAC3E;;;;;;AAkBA,SAAgB,sBACf,SAC0B;CAC1B,IAAI,CAAC,MAAM,QAAQ,OAAO,GACzB,OAAO;CAGR,MAAM,QAAkB,CAAC;CACzB,MAAM,aAAuB,CAAC;CAE9B,KAAK,MAAM,QAAQ,SAAS;EAC3B,IAAI,CAAC,SAAS,IAAI,GACjB;EAED,IAAI,KAAK,SAAS,QAAQ;GACzB,IAAI,OAAO,KAAK,SAAS,UACxB,MAAM,KAAK,KAAK,IAAI;GAErB;EACD;EACA,IAAI,KAAK,SAAS,aAAa;GAG9B,IAAI,OAAO,KAAK,SAAS,UACxB,WAAW,KAAK,KAAK,IAAI;GAE1B,IAAI,MAAM,QAAQ,KAAK,OAAO;SACxB,MAAM,SAAS,KAAK,SACxB,IAAI,SAAS,KAAK,KAAK,OAAO,MAAM,SAAS,UAC5C,WAAW,KAAK,MAAM,IAAI;GAAA;GAI7B,IAAI,MAAM,QAAQ,KAAK,OAAO;SACxB,MAAM,SAAS,KAAK,SACxB,IAAI,SAAS,KAAK,KAAK,OAAO,MAAM,SAAS,UAC5C,WAAW,KAAK,MAAM,IAAI;GAAA;EAI9B;CACD;CAEA,OAAO;EAAE,MAAM,MAAM,KAAK,EAAE;EAAG,WAAW,WAAW,KAAK,IAAI;CAAE;AACjE;;;;;AAMA,SAAgB,wBAAwB,MAAgC;CACvE,IAAI,CAAC,SAAS,IAAI,KAAK,CAAC,MAAM,QAAQ,KAAK,OAAO,GACjD,OAAO;EAAE;EAAM,SAAS;CAAM;CAE/B,IAAI,UAAU;CACd,KAAK,MAAM,UAAU,KAAK,SAAS;EAClC,IAAI,CAAC,SAAS,MAAM,GACnB;EAED,MAAM,UAAU,OAAO;EACvB,IAAI,CAAC,SAAS,OAAO,GACpB;EAED,MAAM,YAAY,sBAAsB,QAAQ,OAAO;EACvD,IAAI,cAAc,MACjB;EAED,QAAQ,UAAU,UAAU;EAC5B,UAAU;EACV,IACC,UAAU,UAAU,SAAS,KAC7B,QAAQ,qBAAqB,QAC7B,QAAQ,aAAa,MAErB,QAAQ,oBAAoB,UAAU;CAExC;CACA,OAAO;EAAE;EAAM;CAAQ;AACxB;;;;;AAMA,SAAgB,mBAAmB,MAAgC;CAClE,IAAI,CAAC,SAAS,IAAI,KAAK,CAAC,MAAM,QAAQ,KAAK,OAAO,GACjD,OAAO;EAAE;EAAM,SAAS;CAAM;CAE/B,IAAI,UAAU;CACd,KAAK,MAAM,UAAU,KAAK,SAAS;EAClC,IAAI,CAAC,SAAS,MAAM,GACnB;EAED,MAAM,QAAQ,OAAO;EACrB,IAAI,CAAC,SAAS,KAAK,GAClB;EAED,MAAM,YAAY,sBAAsB,MAAM,OAAO;EACrD,IAAI,cAAc,MACjB;EAED,UAAU;EAEV,IAAI,UAAU,KAAK,SAAS,GAC3B,MAAM,UAAU,UAAU;OAE1B,MAAM,UAAU,KAAA;EAEjB,IACC,UAAU,UAAU,SAAS,KAC7B,MAAM,qBAAqB,QAC3B,MAAM,aAAa,MAEnB,MAAM,oBAAoB,UAAU;CAEtC;CACA,OAAO;EAAE;EAAM;CAAQ;AACxB;;;;;AAMA,SAAS,qBACR,MAC6B;CAC7B,MAAM,UAAU,IAAI,YAAY;CAChC,MAAM,UAAU,IAAI,YAAY;CAChC,IAAI,SAAS;CAEb,MAAM,eAAe,SAAyB;EAC7C,IAAI,CAAC,KAAK,WAAW,OAAO,GAC3B,OAAO;EAER,MAAM,UAAU,KAAK,MAAM,CAAc,CAAC,CAAC,KAAK;EAChD,IAAI,YAAY,MAAM,YAAY,UACjC,OAAO;EAER,IAAI;EACJ,IAAI;GACH,SAAS,KAAK,MAAM,OAAO;EAC5B,QAAQ;GACP,OAAO;EACR;EACA,MAAM,EAAE,MAAM,YAAY,YAAY,mBAAmB,MAAM;EAE/D,OAAO,UAAU,SAAS,KAAK,UAAU,UAAU,MAAM;CAC1D;CAEA,OAAO,KAAK,YACX,IAAI,gBAAwC;EAC3C,UAAU,OAAO,YAAY;GAC5B,UAAU,QAAQ,OAAO,OAAO,EAAE,QAAQ,KAAK,CAAC;GAChD,MAAM,QAAQ,OAAO,MAAM,IAAI;GAC/B,SAAS,MAAM,IAAI,KAAK;GACxB,KAAK,MAAM,QAAQ,OAClB,WAAW,QACV,QAAQ,OAAO,GAAG,YAAY,IAAI,EAAE,GAAG,CACxC;EAEF;EACA,MAAM,YAAY;GACjB,IAAI,OAAO,SAAS,GACnB,WAAW,QAAQ,QAAQ,OAAO,YAAY,MAAM,CAAC,CAAC;EAExD;CACD,CAAC,CACF;AACD;AAEA,SAAS,eAAe,QAAiB,GAAG,OAA0B;CACrE,MAAM,OAAO,IAAI,QAAQ,MAAM;CAC/B,KAAK,MAAM,QAAQ,OAClB,KAAK,OAAO,IAAI;CAEjB,OAAO;AACR;;;;;;;;;;AAWA,SAAgB,kCACf,WACgB;CAChB,MAAM,UACL,eAAe,OAAO,SAAS,WAAW,MAAM,OAAO,IAAI;CAE5D,OAAO,OAAO,OAAO,SAAS;EAC7B,MAAM,WAAW,MAAM,QAAQ,OAAO,IAAI;EAC1C,IAAI,CAAC,SAAS,MAAM,SAAS,SAAS,MACrC,OAAO;EAER,MAAM,cAAc,SAAS,QAAQ,IAAI,cAAc,KAAK;EAE5D,IAAI,YAAY,SAAS,mBAAmB,GAG3C,OAAO,IAAI,SAAS,qBAAqB,SAAS,IAAI,GAAG;GACxD,QAAQ,SAAS;GACjB,YAAY,SAAS;GAErB,SAAS,eACR,SAAS,SACT,kBACA,kBACD;EACD,CAAC;EAGF,IAAI,YAAY,SAAS,kBAAkB,GAAG;GAE7C,IAAI;GACJ,IAAI;IACH,OAAO,MAAM,SAAS,MAAM,CAAC,CAAC,KAAK;GACpC,QAAQ;IACP,OAAO;GACR;GACA,IAAI;GACJ,IAAI;IACH,SAAS,KAAK,MAAM,IAAI;GACzB,QAAQ;IACP,OAAO;GACR;GACA,MAAM,EAAE,MAAM,YAAY,YACzB,wBAAwB,MAAM;GAC/B,IAAI,CAAC,SAEJ,OAAO;GAER,OAAO,IAAI,SAAS,KAAK,UAAU,UAAU,GAAG;IAC/C,QAAQ,SAAS;IACjB,YAAY,SAAS;IACrB,SAAS,eACR,SAAS,SACT,kBACA,kBACD;GACD,CAAC;EACF;EAEA,OAAO;CACR;AACD"}

package/dist/lib/neon-responses-language-model.d.ts

@@ -5,7 +5,7 @@ import { LanguageModelV3, LanguageModelV3CallOptions, LanguageModelV3GenerateRes
 
 /**
  * Language model for OpenAI models served via the Neon AI Gateway's native
- * Responses API (`/ai-gateway/openai/v1/responses`). This route is required for
+ * Responses API (`/openai/v1/responses`). This route is required for
  * models that are only served natively (e.g. Codex) and unlocks native
  * reasoning and the image-generation tool.
  *

package/dist/lib/neon-responses-language-model.js

@@ -2,7 +2,7 @@ import { OpenAIResponsesLanguageModel } from "@ai-sdk/openai/internal";
 //#region src/lib/neon-responses-language-model.ts
 /**
 * Language model for OpenAI models served via the Neon AI Gateway's native
-* Responses API (`/ai-gateway/openai/v1/responses`). This route is required for
+* Responses API (`/openai/v1/responses`). This route is required for
 * models that are only served natively (e.g. Codex) and unlocks native
 * reasoning and the image-generation tool.
 *

package/dist/lib/neon-responses-language-model.js.map

@@ -1 +1 @@
-{"version":3,"file":"neon-responses-language-model.js","names":[],"sources":["../../src/lib/neon-responses-language-model.ts"],"sourcesContent":["import { OpenAIResponsesLanguageModel } from \"@ai-sdk/openai/internal\";\nimport type {\n\tLanguageModelV3,\n\tLanguageModelV3CallOptions,\n\tLanguageModelV3GenerateResult,\n\tLanguageModelV3StreamResult,\n} from \"@ai-sdk/provider\";\n\n/**\n * Language model for OpenAI models served via the Neon AI Gateway's native\n * Responses API (`/ai-gateway/openai/v1/responses`). This route is required for\n * models that are only served natively (e.g. Codex) and unlocks native\n * reasoning and the image-generation tool.\n *\n * The shared OpenAI Responses model shapes a request based on whether the model\n * is a reasoning model, but its detection is brittle here: the gateway also\n * accepts the legacy `databricks-` prefix, which defeats the upstream bare-id\n * (`gpt-5`) match. For the GPT-5 reasoning family we set the model's own\n * `forceReasoning` provider option so it applies the correct reasoning behavior\n * for both id forms. Users can still override it.\n */\nexport class NeonResponsesLanguageModel\n\textends OpenAIResponsesLanguageModel\n\timplements LanguageModelV3\n{\n\tprivate get isReasoningFamily(): boolean {\n\t\treturn /gpt-5/.test(this.modelId.toLowerCase());\n\t}\n\n\tprivate withForcedReasoning(\n\t\toptions: LanguageModelV3CallOptions,\n\t): LanguageModelV3CallOptions {\n\t\tif (!this.isReasoningFamily) {\n\t\t\treturn options;\n\t\t}\n\t\tconst openai = options.providerOptions?.openai;\n\t\t// Respect an explicit user setting.\n\t\tif (openai != null && \"forceReasoning\" in openai) {\n\t\t\treturn options;\n\t\t}\n\t\treturn {\n\t\t\t...options,\n\t\t\tproviderOptions: {\n\t\t\t\t...options.providerOptions,\n\t\t\t\topenai: { ...openai, forceReasoning: true },\n\t\t\t},\n\t\t};\n\t}\n\n\toverride doGenerate(\n\t\toptions: LanguageModelV3CallOptions,\n\t): Promise<LanguageModelV3GenerateResult> {\n\t\treturn super.doGenerate(this.withForcedReasoning(options));\n\t}\n\n\toverride doStream(\n\t\toptions: LanguageModelV3CallOptions,\n\t): Promise<LanguageModelV3StreamResult> {\n\t\treturn super.doStream(this.withForcedReasoning(options));\n\t}\n}\n"],"mappings":";;;;;;;;;;;;;;;AAqBA,IAAa,6BAAb,cACS,6BAET;CACC,IAAY,oBAA6B;EACxC,OAAO,QAAQ,KAAK,KAAK,QAAQ,YAAY,CAAC;CAC/C;CAEA,oBACC,SAC6B;EAC7B,IAAI,CAAC,KAAK,mBACT,OAAO;EAER,MAAM,SAAS,QAAQ,iBAAiB;EAExC,IAAI,UAAU,QAAQ,oBAAoB,QACzC,OAAO;EAER,OAAO;GACN,GAAG;GACH,iBAAiB;IAChB,GAAG,QAAQ;IACX,QAAQ;KAAE,GAAG;KAAQ,gBAAgB;IAAK;GAC3C;EACD;CACD;CAEA,WACC,SACyC;EACzC,OAAO,MAAM,WAAW,KAAK,oBAAoB,OAAO,CAAC;CAC1D;CAEA,SACC,SACuC;EACvC,OAAO,MAAM,SAAS,KAAK,oBAAoB,OAAO,CAAC;CACxD;AACD"}
+{"version":3,"file":"neon-responses-language-model.js","names":[],"sources":["../../src/lib/neon-responses-language-model.ts"],"sourcesContent":["import { OpenAIResponsesLanguageModel } from \"@ai-sdk/openai/internal\";\nimport type {\n\tLanguageModelV3,\n\tLanguageModelV3CallOptions,\n\tLanguageModelV3GenerateResult,\n\tLanguageModelV3StreamResult,\n} from \"@ai-sdk/provider\";\n\n/**\n * Language model for OpenAI models served via the Neon AI Gateway's native\n * Responses API (`/openai/v1/responses`). This route is required for\n * models that are only served natively (e.g. Codex) and unlocks native\n * reasoning and the image-generation tool.\n *\n * The shared OpenAI Responses model shapes a request based on whether the model\n * is a reasoning model, but its detection is brittle here: the gateway also\n * accepts the legacy `databricks-` prefix, which defeats the upstream bare-id\n * (`gpt-5`) match. For the GPT-5 reasoning family we set the model's own\n * `forceReasoning` provider option so it applies the correct reasoning behavior\n * for both id forms. Users can still override it.\n */\nexport class NeonResponsesLanguageModel\n\textends OpenAIResponsesLanguageModel\n\timplements LanguageModelV3\n{\n\tprivate get isReasoningFamily(): boolean {\n\t\treturn /gpt-5/.test(this.modelId.toLowerCase());\n\t}\n\n\tprivate withForcedReasoning(\n\t\toptions: LanguageModelV3CallOptions,\n\t): LanguageModelV3CallOptions {\n\t\tif (!this.isReasoningFamily) {\n\t\t\treturn options;\n\t\t}\n\t\tconst openai = options.providerOptions?.openai;\n\t\t// Respect an explicit user setting.\n\t\tif (openai != null && \"forceReasoning\" in openai) {\n\t\t\treturn options;\n\t\t}\n\t\treturn {\n\t\t\t...options,\n\t\t\tproviderOptions: {\n\t\t\t\t...options.providerOptions,\n\t\t\t\topenai: { ...openai, forceReasoning: true },\n\t\t\t},\n\t\t};\n\t}\n\n\toverride doGenerate(\n\t\toptions: LanguageModelV3CallOptions,\n\t): Promise<LanguageModelV3GenerateResult> {\n\t\treturn super.doGenerate(this.withForcedReasoning(options));\n\t}\n\n\toverride doStream(\n\t\toptions: LanguageModelV3CallOptions,\n\t): Promise<LanguageModelV3StreamResult> {\n\t\treturn super.doStream(this.withForcedReasoning(options));\n\t}\n}\n"],"mappings":";;;;;;;;;;;;;;;AAqBA,IAAa,6BAAb,cACS,6BAET;CACC,IAAY,oBAA6B;EACxC,OAAO,QAAQ,KAAK,KAAK,QAAQ,YAAY,CAAC;CAC/C;CAEA,oBACC,SAC6B;EAC7B,IAAI,CAAC,KAAK,mBACT,OAAO;EAER,MAAM,SAAS,QAAQ,iBAAiB;EAExC,IAAI,UAAU,QAAQ,oBAAoB,QACzC,OAAO;EAER,OAAO;GACN,GAAG;GACH,iBAAiB;IAChB,GAAG,QAAQ;IACX,QAAQ;KAAE,GAAG;KAAQ,gBAAgB;IAAK;GAC3C;EACD;CACD;CAEA,WACC,SACyC;EACzC,OAAO,MAAM,WAAW,KAAK,oBAAoB,OAAO,CAAC;CAC1D;CAEA,SACC,SACuC;EACvC,OAAO,MAAM,SAAS,KAAK,oBAAoB,OAAO,CAAC;CACxD;AACD"}

package/dist/lib/provider.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"provider.d.ts","names":[],"sources":["../../src/lib/provider.ts"],"mappings":";;;;;;;;cAmCM,iBAAe,CAAA,CAAA;;;;;IAAA,IAAA,eAAA,cAAA,aAAA,CAAA,CAAA;EAAA,CAAA,eAAA,CAAA;AASrB,CAAA,eAAY,CAAA;AAAa,KAAb,aAAA,GAAgB,CAAA,CAAE,KAAL,CAAA,OAAkB,eAAlB,CAAA;AAAkB,UA4C1B,oBAAA,CA5C0B;;AAAR;AA4CnC;;;SAkBS,CAAA,EAAA,MAAA;EAAa;AAGtB;;;QAE6B,CAAA,EAAA,MAAA;;SAGa,CAAA,EAX/B,MAW+B,CAAA,MAAA,EAAA,MAAA,CAAA;;OAGT,CAAA,EAXxB,aAWwB;;AARK,UAArB,YAAA,SAAqB,UAAA,CAAA;EAAU;EAiBhC,CAAA,OAAA,EAfL,eAee,CAAA,EAfG,eAeH;EAAA;eAAU,CAAA,OAAA,EAZZ,eAYY,CAAA,EAZM,eAYN;;EAAwC,IAAA,CAAA,OAAA,EAT7D,eAS6D,CAAA,EAT3C,eAS2C;EA0F/D;gBAhGE,MAAA,CAAe;;;;iBAMd,UAAA,WAAoB,uBAA4B;;cA0FnD,MAAI"}
+{"version":3,"file":"provider.d.ts","names":[],"sources":["../../src/lib/provider.ts"],"mappings":";;;;;;;;cAoCM,iBAAe,CAAA,CAAA;;;;;IAAA,IAAA,eAAA,cAAA,aAAA,CAAA,CAAA;EAAA,CAAA,eAAA,CAAA;AASrB,CAAA,eAAY,CAAA;AAAa,KAAb,aAAA,GAAgB,CAAA,CAAE,KAAL,CAAA,OAAkB,eAAlB,CAAA;AAAkB,UA4C1B,oBAAA,CA5C0B;;AAAR;AA4CnC;;;SAkBS,CAAA,EAAA,MAAA;EAAa;AAGtB;;;QAE6B,CAAA,EAAA,MAAA;;SAGa,CAAA,EAX/B,MAW+B,CAAA,MAAA,EAAA,MAAA,CAAA;;OAGT,CAAA,EAXxB,aAWwB;;AARK,UAArB,YAAA,SAAqB,UAAA,CAAA;EAAU;EAiBhC,CAAA,OAAA,EAfL,eAee,CAAA,EAfG,eAeH;EAAA;eAAU,CAAA,OAAA,EAZZ,eAYY,CAAA,EAZM,eAYN;;EAAwC,IAAA,CAAA,OAAA,EAT7D,eAS6D,CAAA,EAT3C,eAS2C;EA+F/D;gBArGE,MAAA,CAAe;;;;iBAMd,UAAA,WAAoB,uBAA4B;;cA+FnD,MAAI"}

package/dist/lib/provider.js

@@ -2,6 +2,7 @@ import { NeonAnthropicLanguageModel } from "./neon-anthropic-language-model.js";
 import { getNeonModelRoute } from "./neon-model-capabilities.js";
 import { NeonChatLanguageModel } from "./neon-chat-language-model.js";
 import { NeonResponsesLanguageModel } from "./neon-responses-language-model.js";
+import { wrapFetchWithHarmonyNormalization } from "./neon-harmony-normalize.js";
 import { VERSION } from "./version.js";
 import { openai } from "@ai-sdk/openai";
 import { NoSuchModelError } from "@ai-sdk/provider";
@@ -13,7 +14,7 @@ import { z } from "zod/v4";
 *
 * `createNeon()` returns a provider that routes each model to the best gateway
 * endpoint based on its id (Anthropic → native Messages, OpenAI → native
-* Responses incl. Codex, everything else → unified MLflow), so a single
+* Responses incl. Codex, everything else → unified Chat Completions), so a single
 * `neon('claude-...')` call (same base URL + token) reaches the whole catalog.
 * Ids use the canonical Neon (unprefixed) form; the legacy `databricks-` prefix
 * is also accepted. Configure with the branch-scoped `NEON_AI_GATEWAY_BASE_URL` +
@@ -70,23 +71,23 @@ function createNeon(options = {}) {
 	}, `neondatabase/ai-sdk-provider/${VERSION}`);
 	const createAnthropicModel = (modelId) => new NeonAnthropicLanguageModel(modelId, {
 		provider: "neon.anthropic",
-		baseURL: `${getHost()}/ai-gateway/anthropic/v1`,
+		baseURL: `${getHost()}/anthropic/v1`,
 		headers: () => getHeaders({ "anthropic-version": "2023-06-01" }),
 		fetch: options.fetch,
 		generateId
 	});
 	const createOpenAIModel = (modelId) => new NeonResponsesLanguageModel(modelId, {
 		provider: "neon.openai.responses",
-		url: ({ path }) => `${getHost()}/ai-gateway/openai/v1${path}`,
+		url: ({ path }) => `${getHost()}/openai/v1${path}`,
 		headers: getHeaders,
 		fetch: options.fetch,
 		fileIdPrefixes: ["file-"]
 	});
 	const createChatModel = (modelId) => new NeonChatLanguageModel(modelId, {
 		provider: "neon.chat",
-		url: ({ path }) => `${getHost()}/ai-gateway/mlflow/v1${path}`,
+		url: ({ path }) => `${getHost()}/v1${path}`,
 		headers: getHeaders,
-		fetch: options.fetch,
+		fetch: wrapFetchWithHarmonyNormalization(options.fetch),
 		errorStructure: neonErrorStructure,
 		transformRequestBody: transformNeonRequestBody,
 		supportsStructuredOutputs: true

package/dist/lib/provider.js.map

@@ -1 +1 @@
-{"version":3,"file":"provider.js","names":["openaiProvider"],"sources":["../../src/lib/provider.ts"],"sourcesContent":["/**\n * Community Vercel AI SDK provider for the Neon AI Gateway.\n *\n * `createNeon()` returns a provider that routes each model to the best gateway\n * endpoint based on its id (Anthropic → native Messages, OpenAI → native\n * Responses incl. Codex, everything else → unified MLflow), so a single\n * `neon('claude-...')` call (same base URL + token) reaches the whole catalog.\n * Ids use the canonical Neon (unprefixed) form; the legacy `databricks-` prefix\n * is also accepted. Configure with the branch-scoped `NEON_AI_GATEWAY_BASE_URL` +\n * `NEON_AI_GATEWAY_TOKEN` emitted by `neonctl env pull` / `neon dev`, or pass\n * `baseURL` / `apiKey` explicitly.\n */\nimport { openai as openaiProvider } from \"@ai-sdk/openai\";\nimport type { ProviderErrorStructure } from \"@ai-sdk/openai-compatible\";\nimport {\n\ttype LanguageModelV3,\n\tNoSuchModelError,\n\ttype ProviderV3,\n} from \"@ai-sdk/provider\";\nimport {\n\ttype FetchFunction,\n\tgenerateId,\n\tloadApiKey,\n\tloadSetting,\n\twithoutTrailingSlash,\n\twithUserAgentSuffix,\n} from \"@ai-sdk/provider-utils\";\nimport { z } from \"zod/v4\";\nimport { NeonAnthropicLanguageModel } from \"./neon-anthropic-language-model.js\";\nimport { NeonChatLanguageModel } from \"./neon-chat-language-model.js\";\nimport type { NeonChatModelId } from \"./neon-chat-options.js\";\nimport { getNeonModelRoute } from \"./neon-model-capabilities.js\";\nimport { NeonResponsesLanguageModel } from \"./neon-responses-language-model.js\";\nimport { VERSION } from \"./version.js\";\n\nconst neonErrorSchema = z.object({\n\terror: z.object({\n\t\tmessage: z.string(),\n\t\ttype: z.string().nullish(),\n\t\tparam: z.unknown().nullish(),\n\t\tcode: z.unknown().nullish(),\n\t}),\n});\n\nexport type NeonErrorData = z.infer<typeof neonErrorSchema>;\n\nconst neonErrorStructure: ProviderErrorStructure<NeonErrorData> = {\n\terrorSchema: neonErrorSchema,\n\terrorToMessage: (data) => data.error.message,\n};\n\n/**\n * Recursively remove the JSON Schema `$schema` marker, which some gateway\n * backends (notably Gemini) reject in tool/structured-output schemas. Other\n * backends ignore its absence, so stripping it everywhere is safe.\n */\nfunction stripJsonSchemaMarker(value: unknown): unknown {\n\tif (Array.isArray(value)) {\n\t\treturn value.map(stripJsonSchemaMarker);\n\t}\n\tif (value !== null && typeof value === \"object\") {\n\t\tconst result: Record<string, unknown> = {};\n\t\tfor (const [key, entry] of Object.entries(value)) {\n\t\t\tif (key === \"$schema\") {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tresult[key] = stripJsonSchemaMarker(entry);\n\t\t}\n\t\treturn result;\n\t}\n\treturn value;\n}\n\nfunction transformNeonRequestBody(\n\targs: Record<string, unknown>,\n): Record<string, unknown> {\n\tconst transformed = { ...args };\n\tif (transformed.tools != null) {\n\t\ttransformed.tools = stripJsonSchemaMarker(transformed.tools);\n\t}\n\tif (transformed.response_format != null) {\n\t\ttransformed.response_format = stripJsonSchemaMarker(\n\t\t\ttransformed.response_format,\n\t\t);\n\t}\n\treturn transformed;\n}\n\nexport interface NeonProviderSettings {\n\t/**\n\t * Neon AI Gateway base URL — the branch-scoped host root, e.g.\n\t * `https://<branch-id>-api.ai.<region>.aws.neon.tech`. The gateway paths are\n\t * appended internally. Falls back to the `NEON_AI_GATEWAY_BASE_URL` env var.\n\t */\n\tbaseURL?: string;\n\n\t/**\n\t * Neon AI Gateway platform token (the `nt_live_...` value). Falls back to the\n\t * `NEON_AI_GATEWAY_TOKEN` env var.\n\t */\n\tapiKey?: string;\n\n\t/** Custom headers to include in the requests. */\n\theaders?: Record<string, string>;\n\n\t/** Custom fetch implementation (e.g. for testing or middleware). */\n\tfetch?: FetchFunction;\n}\n\nexport interface NeonProvider extends ProviderV3 {\n\t/** Creates a Neon AI Gateway model for text generation. */\n\t(modelId: NeonChatModelId): LanguageModelV3;\n\n\t/** Creates a Neon AI Gateway model for text generation. */\n\tlanguageModel(modelId: NeonChatModelId): LanguageModelV3;\n\n\t/** Creates a Neon AI Gateway chat model for text generation. */\n\tchat(modelId: NeonChatModelId): LanguageModelV3;\n\n\t/** OpenAI Responses tools (e.g. `imageGeneration`) for OpenAI-routed models. */\n\ttools: typeof openaiProvider.tools;\n\n\t/** @deprecated Use `embeddingModel` instead. */\n\ttextEmbeddingModel(modelId: string): never;\n}\n\nexport function createNeon(options: NeonProviderSettings = {}): NeonProvider {\n\t// Resolved lazily so that `createNeon()` and the default `neon` instance do\n\t// not throw at import time when configuration comes from the environment.\n\tconst getHost = () =>\n\t\twithoutTrailingSlash(\n\t\t\tloadSetting({\n\t\t\t\tsettingValue: options.baseURL,\n\t\t\t\tenvironmentVariableName: \"NEON_AI_GATEWAY_BASE_URL\",\n\t\t\t\tsettingName: \"baseURL\",\n\t\t\t\tdescription: \"Neon AI Gateway base URL\",\n\t\t\t}),\n\t\t);\n\n\tconst getHeaders = (extra?: Record<string, string>) =>\n\t\twithUserAgentSuffix(\n\t\t\t{\n\t\t\t\tAuthorization: `Bearer ${loadApiKey({\n\t\t\t\t\tapiKey: options.apiKey,\n\t\t\t\t\tenvironmentVariableName: \"NEON_AI_GATEWAY_TOKEN\",\n\t\t\t\t\tdescription: \"Neon AI Gateway token\",\n\t\t\t\t})}`,\n\t\t\t\t...extra,\n\t\t\t\t...options.headers,\n\t\t\t},\n\t\t\t`neondatabase/ai-sdk-provider/${VERSION}`,\n\t\t);\n\n\t// Anthropic models -> native Messages API.\n\tconst createAnthropicModel = (modelId: NeonChatModelId) =>\n\t\tnew NeonAnthropicLanguageModel(modelId, {\n\t\t\tprovider: \"neon.anthropic\",\n\t\t\tbaseURL: `${getHost()}/ai-gateway/anthropic/v1`,\n\t\t\theaders: () => getHeaders({ \"anthropic-version\": \"2023-06-01\" }),\n\t\t\tfetch: options.fetch,\n\t\t\tgenerateId,\n\t\t});\n\n\t// OpenAI models (incl. Codex, only served natively) -> Responses API.\n\tconst createOpenAIModel = (modelId: NeonChatModelId) =>\n\t\tnew NeonResponsesLanguageModel(modelId, {\n\t\t\tprovider: \"neon.openai.responses\",\n\t\t\turl: ({ path }) => `${getHost()}/ai-gateway/openai/v1${path}`,\n\t\t\theaders: getHeaders,\n\t\t\tfetch: options.fetch,\n\t\t\tfileIdPrefixes: [\"file-\"],\n\t\t});\n\n\t// Everything else (Gemini, Llama, Qwen, gpt-oss, ...) -> unified MLflow\n\t// endpoint. Gemini is here because its native endpoint can't stream.\n\tconst createChatModel = (modelId: NeonChatModelId) =>\n\t\tnew NeonChatLanguageModel(modelId, {\n\t\t\tprovider: \"neon.chat\",\n\t\t\turl: ({ path }) => `${getHost()}/ai-gateway/mlflow/v1${path}`,\n\t\t\theaders: getHeaders,\n\t\t\tfetch: options.fetch,\n\t\t\terrorStructure: neonErrorStructure,\n\t\t\ttransformRequestBody: transformNeonRequestBody,\n\t\t\tsupportsStructuredOutputs: true,\n\t\t});\n\n\tconst createLanguageModel = (modelId: NeonChatModelId): LanguageModelV3 => {\n\t\tswitch (getNeonModelRoute(modelId)) {\n\t\t\tcase \"anthropic\":\n\t\t\t\treturn createAnthropicModel(modelId);\n\t\t\tcase \"openai\":\n\t\t\t\treturn createOpenAIModel(modelId);\n\t\t\tdefault:\n\t\t\t\treturn createChatModel(modelId);\n\t\t}\n\t};\n\n\tconst provider = (modelId: NeonChatModelId) => createLanguageModel(modelId);\n\n\tprovider.specificationVersion = \"v3\" as const;\n\tprovider.languageModel = createLanguageModel;\n\tprovider.chat = createLanguageModel;\n\tprovider.tools = openaiProvider.tools;\n\n\tprovider.embeddingModel = (modelId: string) => {\n\t\tthrow new NoSuchModelError({ modelId, modelType: \"embeddingModel\" });\n\t};\n\tprovider.textEmbeddingModel = provider.embeddingModel;\n\tprovider.imageModel = (modelId: string) => {\n\t\tthrow new NoSuchModelError({ modelId, modelType: \"imageModel\" });\n\t};\n\n\treturn provider;\n}\n\n/** Default Neon provider instance (reads `NEON_AI_GATEWAY_*` env vars). */\nexport const neon = createNeon();\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;AA8CA,MAAM,qBAA4D;CACjE,aAZuB,EAAE,OAAO,EAChC,OAAO,EAAE,OAAO;EACf,SAAS,EAAE,OAAO;EAClB,MAAM,EAAE,OAAO,CAAC,CAAC,QAAQ;EACzB,OAAO,EAAE,QAAQ,CAAC,CAAC,QAAQ;EAC3B,MAAM,EAAE,QAAQ,CAAC,CAAC,QAAQ;CAC3B,CAAC,EACF,CAK4B;CAC3B,iBAAiB,SAAS,KAAK,MAAM;AACtC;;;;;;AAOA,SAAS,sBAAsB,OAAyB;CACvD,IAAI,MAAM,QAAQ,KAAK,GACtB,OAAO,MAAM,IAAI,qBAAqB;CAEvC,IAAI,UAAU,QAAQ,OAAO,UAAU,UAAU;EAChD,MAAM,SAAkC,CAAC;EACzC,KAAK,MAAM,CAAC,KAAK,UAAU,OAAO,QAAQ,KAAK,GAAG;GACjD,IAAI,QAAQ,WACX;GAED,OAAO,OAAO,sBAAsB,KAAK;EAC1C;EACA,OAAO;CACR;CACA,OAAO;AACR;AAEA,SAAS,yBACR,MAC0B;CAC1B,MAAM,cAAc,EAAE,GAAG,KAAK;CAC9B,IAAI,YAAY,SAAS,MACxB,YAAY,QAAQ,sBAAsB,YAAY,KAAK;CAE5D,IAAI,YAAY,mBAAmB,MAClC,YAAY,kBAAkB,sBAC7B,YAAY,eACb;CAED,OAAO;AACR;AAwCA,SAAgB,WAAW,UAAgC,CAAC,GAAiB;CAG5E,MAAM,gBACL,qBACC,YAAY;EACX,cAAc,QAAQ;EACtB,yBAAyB;EACzB,aAAa;EACb,aAAa;CACd,CAAC,CACF;CAED,MAAM,cAAc,UACnB,oBACC;EACC,eAAe,UAAU,WAAW;GACnC,QAAQ,QAAQ;GAChB,yBAAyB;GACzB,aAAa;EACd,CAAC;EACD,GAAG;EACH,GAAG,QAAQ;CACZ,GACA,gCAAgC,SACjC;CAGD,MAAM,wBAAwB,YAC7B,IAAI,2BAA2B,SAAS;EACvC,UAAU;EACV,SAAS,GAAG,QAAQ,EAAE;EACtB,eAAe,WAAW,EAAE,qBAAqB,aAAa,CAAC;EAC/D,OAAO,QAAQ;EACf;CACD,CAAC;CAGF,MAAM,qBAAqB,YAC1B,IAAI,2BAA2B,SAAS;EACvC,UAAU;EACV,MAAM,EAAE,WAAW,GAAG,QAAQ,EAAE,uBAAuB;EACvD,SAAS;EACT,OAAO,QAAQ;EACf,gBAAgB,CAAC,OAAO;CACzB,CAAC;CAIF,MAAM,mBAAmB,YACxB,IAAI,sBAAsB,SAAS;EAClC,UAAU;EACV,MAAM,EAAE,WAAW,GAAG,QAAQ,EAAE,uBAAuB;EACvD,SAAS;EACT,OAAO,QAAQ;EACf,gBAAgB;EAChB,sBAAsB;EACtB,2BAA2B;CAC5B,CAAC;CAEF,MAAM,uBAAuB,YAA8C;EAC1E,QAAQ,kBAAkB,OAAO,GAAjC;GACC,KAAK,aACJ,OAAO,qBAAqB,OAAO;GACpC,KAAK,UACJ,OAAO,kBAAkB,OAAO;GACjC,SACC,OAAO,gBAAgB,OAAO;EAChC;CACD;CAEA,MAAM,YAAY,YAA6B,oBAAoB,OAAO;CAE1E,SAAS,uBAAuB;CAChC,SAAS,gBAAgB;CACzB,SAAS,OAAO;CAChB,SAAS,QAAQA,OAAe;CAEhC,SAAS,kBAAkB,YAAoB;EAC9C,MAAM,IAAI,iBAAiB;GAAE;GAAS,WAAW;EAAiB,CAAC;CACpE;CACA,SAAS,qBAAqB,SAAS;CACvC,SAAS,cAAc,YAAoB;EAC1C,MAAM,IAAI,iBAAiB;GAAE;GAAS,WAAW;EAAa,CAAC;CAChE;CAEA,OAAO;AACR;;AAGA,MAAa,OAAO,WAAW"}
+{"version":3,"file":"provider.js","names":["openaiProvider"],"sources":["../../src/lib/provider.ts"],"sourcesContent":["/**\n * Community Vercel AI SDK provider for the Neon AI Gateway.\n *\n * `createNeon()` returns a provider that routes each model to the best gateway\n * endpoint based on its id (Anthropic → native Messages, OpenAI → native\n * Responses incl. Codex, everything else → unified Chat Completions), so a single\n * `neon('claude-...')` call (same base URL + token) reaches the whole catalog.\n * Ids use the canonical Neon (unprefixed) form; the legacy `databricks-` prefix\n * is also accepted. Configure with the branch-scoped `NEON_AI_GATEWAY_BASE_URL` +\n * `NEON_AI_GATEWAY_TOKEN` emitted by `neonctl env pull` / `neon dev`, or pass\n * `baseURL` / `apiKey` explicitly.\n */\nimport { openai as openaiProvider } from \"@ai-sdk/openai\";\nimport type { ProviderErrorStructure } from \"@ai-sdk/openai-compatible\";\nimport {\n\ttype LanguageModelV3,\n\tNoSuchModelError,\n\ttype ProviderV3,\n} from \"@ai-sdk/provider\";\nimport {\n\ttype FetchFunction,\n\tgenerateId,\n\tloadApiKey,\n\tloadSetting,\n\twithoutTrailingSlash,\n\twithUserAgentSuffix,\n} from \"@ai-sdk/provider-utils\";\nimport { z } from \"zod/v4\";\nimport { NeonAnthropicLanguageModel } from \"./neon-anthropic-language-model.js\";\nimport { NeonChatLanguageModel } from \"./neon-chat-language-model.js\";\nimport type { NeonChatModelId } from \"./neon-chat-options.js\";\nimport { wrapFetchWithHarmonyNormalization } from \"./neon-harmony-normalize.js\";\nimport { getNeonModelRoute } from \"./neon-model-capabilities.js\";\nimport { NeonResponsesLanguageModel } from \"./neon-responses-language-model.js\";\nimport { VERSION } from \"./version.js\";\n\nconst neonErrorSchema = z.object({\n\terror: z.object({\n\t\tmessage: z.string(),\n\t\ttype: z.string().nullish(),\n\t\tparam: z.unknown().nullish(),\n\t\tcode: z.unknown().nullish(),\n\t}),\n});\n\nexport type NeonErrorData = z.infer<typeof neonErrorSchema>;\n\nconst neonErrorStructure: ProviderErrorStructure<NeonErrorData> = {\n\terrorSchema: neonErrorSchema,\n\terrorToMessage: (data) => data.error.message,\n};\n\n/**\n * Recursively remove the JSON Schema `$schema` marker, which some gateway\n * backends (notably Gemini) reject in tool/structured-output schemas. Other\n * backends ignore its absence, so stripping it everywhere is safe.\n */\nfunction stripJsonSchemaMarker(value: unknown): unknown {\n\tif (Array.isArray(value)) {\n\t\treturn value.map(stripJsonSchemaMarker);\n\t}\n\tif (value !== null && typeof value === \"object\") {\n\t\tconst result: Record<string, unknown> = {};\n\t\tfor (const [key, entry] of Object.entries(value)) {\n\t\t\tif (key === \"$schema\") {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tresult[key] = stripJsonSchemaMarker(entry);\n\t\t}\n\t\treturn result;\n\t}\n\treturn value;\n}\n\nfunction transformNeonRequestBody(\n\targs: Record<string, unknown>,\n): Record<string, unknown> {\n\tconst transformed = { ...args };\n\tif (transformed.tools != null) {\n\t\ttransformed.tools = stripJsonSchemaMarker(transformed.tools);\n\t}\n\tif (transformed.response_format != null) {\n\t\ttransformed.response_format = stripJsonSchemaMarker(\n\t\t\ttransformed.response_format,\n\t\t);\n\t}\n\treturn transformed;\n}\n\nexport interface NeonProviderSettings {\n\t/**\n\t * Neon AI Gateway base URL — the branch-scoped host root, e.g.\n\t * `https://<branch-id>-api.ai.<region>.aws.neon.tech`. The gateway paths are\n\t * appended internally. Falls back to the `NEON_AI_GATEWAY_BASE_URL` env var.\n\t */\n\tbaseURL?: string;\n\n\t/**\n\t * Neon AI Gateway platform token (the `nt_live_...` value). Falls back to the\n\t * `NEON_AI_GATEWAY_TOKEN` env var.\n\t */\n\tapiKey?: string;\n\n\t/** Custom headers to include in the requests. */\n\theaders?: Record<string, string>;\n\n\t/** Custom fetch implementation (e.g. for testing or middleware). */\n\tfetch?: FetchFunction;\n}\n\nexport interface NeonProvider extends ProviderV3 {\n\t/** Creates a Neon AI Gateway model for text generation. */\n\t(modelId: NeonChatModelId): LanguageModelV3;\n\n\t/** Creates a Neon AI Gateway model for text generation. */\n\tlanguageModel(modelId: NeonChatModelId): LanguageModelV3;\n\n\t/** Creates a Neon AI Gateway chat model for text generation. */\n\tchat(modelId: NeonChatModelId): LanguageModelV3;\n\n\t/** OpenAI Responses tools (e.g. `imageGeneration`) for OpenAI-routed models. */\n\ttools: typeof openaiProvider.tools;\n\n\t/** @deprecated Use `embeddingModel` instead. */\n\ttextEmbeddingModel(modelId: string): never;\n}\n\nexport function createNeon(options: NeonProviderSettings = {}): NeonProvider {\n\t// Resolved lazily so that `createNeon()` and the default `neon` instance do\n\t// not throw at import time when configuration comes from the environment.\n\tconst getHost = () =>\n\t\twithoutTrailingSlash(\n\t\t\tloadSetting({\n\t\t\t\tsettingValue: options.baseURL,\n\t\t\t\tenvironmentVariableName: \"NEON_AI_GATEWAY_BASE_URL\",\n\t\t\t\tsettingName: \"baseURL\",\n\t\t\t\tdescription: \"Neon AI Gateway base URL\",\n\t\t\t}),\n\t\t);\n\n\tconst getHeaders = (extra?: Record<string, string>) =>\n\t\twithUserAgentSuffix(\n\t\t\t{\n\t\t\t\tAuthorization: `Bearer ${loadApiKey({\n\t\t\t\t\tapiKey: options.apiKey,\n\t\t\t\t\tenvironmentVariableName: \"NEON_AI_GATEWAY_TOKEN\",\n\t\t\t\t\tdescription: \"Neon AI Gateway token\",\n\t\t\t\t})}`,\n\t\t\t\t...extra,\n\t\t\t\t...options.headers,\n\t\t\t},\n\t\t\t`neondatabase/ai-sdk-provider/${VERSION}`,\n\t\t);\n\n\t// Anthropic models -> native Messages API.\n\tconst createAnthropicModel = (modelId: NeonChatModelId) =>\n\t\tnew NeonAnthropicLanguageModel(modelId, {\n\t\t\tprovider: \"neon.anthropic\",\n\t\t\tbaseURL: `${getHost()}/anthropic/v1`,\n\t\t\theaders: () => getHeaders({ \"anthropic-version\": \"2023-06-01\" }),\n\t\t\tfetch: options.fetch,\n\t\t\tgenerateId,\n\t\t});\n\n\t// OpenAI models (incl. Codex, only served natively) -> Responses API.\n\tconst createOpenAIModel = (modelId: NeonChatModelId) =>\n\t\tnew NeonResponsesLanguageModel(modelId, {\n\t\t\tprovider: \"neon.openai.responses\",\n\t\t\turl: ({ path }) => `${getHost()}/openai/v1${path}`,\n\t\t\theaders: getHeaders,\n\t\t\tfetch: options.fetch,\n\t\t\tfileIdPrefixes: [\"file-\"],\n\t\t});\n\n\t// Everything else (Gemini, Llama, Qwen, gpt-oss, ...) -> unified Chat\n\t// Completions endpoint. Gemini is here because its native endpoint can't stream.\n\tconst createChatModel = (modelId: NeonChatModelId) =>\n\t\tnew NeonChatLanguageModel(modelId, {\n\t\t\tprovider: \"neon.chat\",\n\t\t\turl: ({ path }) => `${getHost()}/v1${path}`,\n\t\t\theaders: getHeaders,\n\t\t\t// TODO(#308): Temporary workaround for gpt-oss. The gateway returns a\n\t\t\t// non-compliant \"harmony\" content-array shape; this wrapper normalizes it\n\t\t\t// to the Chat Completions contract and is a transparent pass-through for\n\t\t\t// every compliant response. Remove it (and neon-harmony-normalize.ts) once\n\t\t\t// the gateway returns spec-compliant gpt-oss responses.\n\t\t\tfetch: wrapFetchWithHarmonyNormalization(options.fetch),\n\t\t\terrorStructure: neonErrorStructure,\n\t\t\ttransformRequestBody: transformNeonRequestBody,\n\t\t\tsupportsStructuredOutputs: true,\n\t\t});\n\n\tconst createLanguageModel = (modelId: NeonChatModelId): LanguageModelV3 => {\n\t\tswitch (getNeonModelRoute(modelId)) {\n\t\t\tcase \"anthropic\":\n\t\t\t\treturn createAnthropicModel(modelId);\n\t\t\tcase \"openai\":\n\t\t\t\treturn createOpenAIModel(modelId);\n\t\t\tdefault:\n\t\t\t\treturn createChatModel(modelId);\n\t\t}\n\t};\n\n\tconst provider = (modelId: NeonChatModelId) => createLanguageModel(modelId);\n\n\tprovider.specificationVersion = \"v3\" as const;\n\tprovider.languageModel = createLanguageModel;\n\tprovider.chat = createLanguageModel;\n\tprovider.tools = openaiProvider.tools;\n\n\tprovider.embeddingModel = (modelId: string) => {\n\t\tthrow new NoSuchModelError({ modelId, modelType: \"embeddingModel\" });\n\t};\n\tprovider.textEmbeddingModel = provider.embeddingModel;\n\tprovider.imageModel = (modelId: string) => {\n\t\tthrow new NoSuchModelError({ modelId, modelType: \"imageModel\" });\n\t};\n\n\treturn provider;\n}\n\n/** Default Neon provider instance (reads `NEON_AI_GATEWAY_*` env vars). */\nexport const neon = createNeon();\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;AA+CA,MAAM,qBAA4D;CACjE,aAZuB,EAAE,OAAO,EAChC,OAAO,EAAE,OAAO;EACf,SAAS,EAAE,OAAO;EAClB,MAAM,EAAE,OAAO,CAAC,CAAC,QAAQ;EACzB,OAAO,EAAE,QAAQ,CAAC,CAAC,QAAQ;EAC3B,MAAM,EAAE,QAAQ,CAAC,CAAC,QAAQ;CAC3B,CAAC,EACF,CAK4B;CAC3B,iBAAiB,SAAS,KAAK,MAAM;AACtC;;;;;;AAOA,SAAS,sBAAsB,OAAyB;CACvD,IAAI,MAAM,QAAQ,KAAK,GACtB,OAAO,MAAM,IAAI,qBAAqB;CAEvC,IAAI,UAAU,QAAQ,OAAO,UAAU,UAAU;EAChD,MAAM,SAAkC,CAAC;EACzC,KAAK,MAAM,CAAC,KAAK,UAAU,OAAO,QAAQ,KAAK,GAAG;GACjD,IAAI,QAAQ,WACX;GAED,OAAO,OAAO,sBAAsB,KAAK;EAC1C;EACA,OAAO;CACR;CACA,OAAO;AACR;AAEA,SAAS,yBACR,MAC0B;CAC1B,MAAM,cAAc,EAAE,GAAG,KAAK;CAC9B,IAAI,YAAY,SAAS,MACxB,YAAY,QAAQ,sBAAsB,YAAY,KAAK;CAE5D,IAAI,YAAY,mBAAmB,MAClC,YAAY,kBAAkB,sBAC7B,YAAY,eACb;CAED,OAAO;AACR;AAwCA,SAAgB,WAAW,UAAgC,CAAC,GAAiB;CAG5E,MAAM,gBACL,qBACC,YAAY;EACX,cAAc,QAAQ;EACtB,yBAAyB;EACzB,aAAa;EACb,aAAa;CACd,CAAC,CACF;CAED,MAAM,cAAc,UACnB,oBACC;EACC,eAAe,UAAU,WAAW;GACnC,QAAQ,QAAQ;GAChB,yBAAyB;GACzB,aAAa;EACd,CAAC;EACD,GAAG;EACH,GAAG,QAAQ;CACZ,GACA,gCAAgC,SACjC;CAGD,MAAM,wBAAwB,YAC7B,IAAI,2BAA2B,SAAS;EACvC,UAAU;EACV,SAAS,GAAG,QAAQ,EAAE;EACtB,eAAe,WAAW,EAAE,qBAAqB,aAAa,CAAC;EAC/D,OAAO,QAAQ;EACf;CACD,CAAC;CAGF,MAAM,qBAAqB,YAC1B,IAAI,2BAA2B,SAAS;EACvC,UAAU;EACV,MAAM,EAAE,WAAW,GAAG,QAAQ,EAAE,YAAY;EAC5C,SAAS;EACT,OAAO,QAAQ;EACf,gBAAgB,CAAC,OAAO;CACzB,CAAC;CAIF,MAAM,mBAAmB,YACxB,IAAI,sBAAsB,SAAS;EAClC,UAAU;EACV,MAAM,EAAE,WAAW,GAAG,QAAQ,EAAE,KAAK;EACrC,SAAS;EAMT,OAAO,kCAAkC,QAAQ,KAAK;EACtD,gBAAgB;EAChB,sBAAsB;EACtB,2BAA2B;CAC5B,CAAC;CAEF,MAAM,uBAAuB,YAA8C;EAC1E,QAAQ,kBAAkB,OAAO,GAAjC;GACC,KAAK,aACJ,OAAO,qBAAqB,OAAO;GACpC,KAAK,UACJ,OAAO,kBAAkB,OAAO;GACjC,SACC,OAAO,gBAAgB,OAAO;EAChC;CACD;CAEA,MAAM,YAAY,YAA6B,oBAAoB,OAAO;CAE1E,SAAS,uBAAuB;CAChC,SAAS,gBAAgB;CACzB,SAAS,OAAO;CAChB,SAAS,QAAQA,OAAe;CAEhC,SAAS,kBAAkB,YAAoB;EAC9C,MAAM,IAAI,iBAAiB;GAAE;GAAS,WAAW;EAAiB,CAAC;CACpE;CACA,SAAS,qBAAqB,SAAS;CACvC,SAAS,cAAc,YAAoB;EAC1C,MAAM,IAAI,iBAAiB;GAAE;GAAS,WAAW;EAAa,CAAC;CAChE;CAEA,OAAO;AACR;;AAGA,MAAa,OAAO,WAAW"}

package/dist/package.js

@@ -1,5 +1,5 @@
 //#region package.json
-var version = "0.6.0";
+var version = "0.7.1";
 //#endregion
 export { version };
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@neon/ai-sdk-provider",
-  "version": "0.6.0",
+  "version": "0.7.1",
   "description": "Community Vercel AI SDK provider for the Neon AI Gateway.",
   "keywords": [
     "neon",
@@ -38,11 +38,11 @@
   ],
   "devDependencies": {
     "@types/node": "^20.19.0",
-    "@vitest/coverage-v8": "3.0.9",
+    "@vitest/coverage-v8": "^3.0.9",
     "ai": "6.0.132",
     "console-fail-test": "0.5.0",
     "tsdown": "^0.14.1",
-    "typescript": "^5.8.2",
+    "typescript": "^5.9.0",
     "vitest": "^3.0.9",
     "zod": "^4.4.3"
   },