@nirmos/sdk 0.3.1 → 0.3.3

package/src/core/transport.ts

@@ -47,7 +47,7 @@ export interface TransportResponse<T> {
   response: Response;
 }
 
-const SDK_VERSION = "0.3.0";
+import { SDK_VERSION } from "./version.js";
 
 export class HttpTransport {
   readonly #config: TransportConfig;
@@ -95,7 +95,7 @@ export class HttpTransport {
         if (
           retryableMethod &&
           attempt < maxRetries &&
-          isRetryableStatus(response.status)
+          (await mayRetryResponse(response))
         ) {
           await response.body?.cancel();
           await waitBeforeRetry(response, attempt, request.options?.signal);
@@ -319,6 +319,18 @@ function isRetryableStatus(status: number): boolean {
   return status === 408 || status === 429 || status >= 500;
 }
 
+/** Server retryability overrides status defaults, including daily provider quota. */
+async function mayRetryResponse(response: Response): Promise<boolean> {
+  if (!isRetryableStatus(response.status)) return false;
+  try {
+    const payload = asUnknownRecord(await response.clone().json());
+    return readBoolean(asUnknownRecord(payload.error), "retryable") !== false;
+  } catch {
+    // Non-JSON legacy responses retain bounded status-based retry behavior.
+    return true;
+  }
+}
+
 function parseRetryAfter(value: string | null): number | undefined {
   if (!value) return undefined;
   const seconds = Number(value);

package/src/core/version.ts

@@ -0,0 +1,2 @@
+// Generated from package.json by scripts/generate-version.mjs.
+export const SDK_VERSION = "0.3.3";

package/src/gateway/chat-stream.ts

@@ -8,7 +8,7 @@ import {
   readString,
 } from "../core/metadata.js";
 import type { RequestMetadata, Usage } from "../core/types.js";
-import { normalizeToolCall, normalizeUsage } from "./mappers.js";
+import { mergeReasoningDetails, normalizeToolCall, normalizeUsage } from "./mappers.js";
 import type {
   ChatChoice,
   ChatCompletion,
@@ -21,6 +21,8 @@ type MutableChoice = {
   index: number;
   role: ChatMessage["role"];
   content: string;
+  reasoning?: string;
+  reasoningDetails?: ChatMessage["reasoningDetails"];
   finishReason: ChatChoice["finishReason"];
   toolCalls: Map<number, FunctionToolCall>;
 };
@@ -140,6 +142,13 @@ export class ChatStream implements AsyncIterable<ChatStreamEvent> {
                   : finishReason;
           }
 
+          const reasoning = readString(delta, "reasoning");
+          if (reasoning) {
+            mutable.reasoning = (mutable.reasoning ?? "") + reasoning;
+            yield { type: "reasoning.delta", id: this.#id, model: this.#model, choiceIndex: index, delta: reasoning, metadata: this.#metadata };
+          }
+          const details = readArray(delta, "reasoning_details", "reasoningDetails") as NonNullable<ChatMessage["reasoningDetails"]>;
+          if (details.length) mutable.reasoningDetails = mergeReasoningDetails(mutable.reasoningDetails ?? [], details);
           const content = readString(delta, "content");
           if (content) {
             mutable.content += content;
@@ -235,6 +244,8 @@ export class ChatStream implements AsyncIterable<ChatStreamEvent> {
         message: {
           role: choice.role,
           content: choice.content,
+          ...(choice.reasoning !== undefined ? { reasoning: choice.reasoning } : {}),
+          ...(choice.reasoningDetails ? { reasoningDetails: choice.reasoningDetails } : {}),
           ...(choice.toolCalls.size > 0
             ? { toolCalls: [...choice.toolCalls.values()] }
             : {}),

package/src/gateway/chat.ts

@@ -77,6 +77,7 @@ export class Chat {
       seed: params.seed,
       user: params.user,
       provider: params.provider,
+      provider_options: params.providerOptions,
       fallback_models: params.fallbackModels,
       routing_strategy: toWireRoutingStrategy(params.routingStrategy),
       metadata: params.metadata,

package/src/gateway/embeddings.ts

@@ -28,6 +28,7 @@ export class Embeddings {
         encoding_format: params.encoding,
         user: params.user,
         provider: params.provider,
+      provider_options: params.providerOptions,
         fallback_models: params.fallbackModels,
         routing_strategy: toWireRoutingStrategy(params.routingStrategy),
       },

package/src/gateway/mappers.ts

@@ -36,6 +36,12 @@ export function toWireMessage(message: ChatMessage): Record<string, unknown> {
               audio: { data: part.data, format: part.format },
             };
           }),
+    ...(message.reasoning !== undefined
+      ? { reasoning: message.reasoning }
+      : {}),
+    ...(message.reasoningDetails?.length
+      ? { reasoning_details: message.reasoningDetails }
+      : {}),
     ...(message.name ? { name: message.name } : {}),
     ...(message.toolCallId ? { tool_call_id: message.toolCallId } : {}),
     ...(message.toolCalls && message.toolCalls.length > 0
@@ -68,8 +74,15 @@ export function normalizeUsage(value: unknown): Usage {
     totalTokens:
       readNumber(usage, "total_tokens", "totalTokens") ??
       inputTokens + outputTokens,
-    cachedTokens: readNumber(usage, "cached_tokens", "cachedTokens"),
-    reasoningTokens: readNumber(usage, "reasoning_tokens", "reasoningTokens"),
+    cachedTokens:
+      readNumber(usage, "cached_tokens", "cachedTokens") ??
+      readNumber(asUnknownRecord(usage.prompt_tokens_details), "cached_tokens"),
+    reasoningTokens:
+      readNumber(usage, "reasoning_tokens", "reasoningTokens") ??
+      readNumber(
+        asUnknownRecord(usage.completion_tokens_details),
+        "reasoning_tokens",
+      ),
   };
 }
 
@@ -183,6 +196,12 @@ export function normalizeMessage(value: unknown): ChatMessage {
         ? content
         : "",
     name: readString(message, "name"),
+    reasoning: readString(message, "reasoning"),
+    reasoningDetails: readArray(
+      message,
+      "reasoning_details",
+      "reasoningDetails",
+    ) as ChatMessage["reasoningDetails"],
     toolCallId: readString(message, "tool_call_id", "toolCallId"),
     toolCalls: toolCalls.length > 0 ? toolCalls : undefined,
   };
@@ -232,3 +251,37 @@ function normalizeFinishReason(
   if (value === "content_filter") return "contentFilter";
   return value;
 }
+
+/** Assemble indexed reasoning blocks without losing encrypted continuation data. */
+export function mergeReasoningDetails(
+  current: NonNullable<ChatMessage["reasoningDetails"]>,
+  incoming: NonNullable<ChatMessage["reasoningDetails"]>,
+): NonNullable<ChatMessage["reasoningDetails"]> {
+  const result = [...current];
+  for (const value of incoming) {
+    const detail = asUnknownRecord(value);
+    const position = result.findIndex((item) => {
+      const previous = asUnknownRecord(item);
+      return (
+        previous.type === detail.type &&
+        ((detail.index !== undefined && previous.index === detail.index) ||
+          (typeof detail.id === "string" && previous.id === detail.id))
+      );
+    });
+    if (position < 0) {
+      result.push(value);
+      continue;
+    }
+    const previous = asUnknownRecord(result[position]);
+    const merged = { ...previous, ...detail };
+    for (const field of ["text", "summary", "data", "signature"]) {
+      if (
+        typeof previous[field] === "string" &&
+        typeof detail[field] === "string"
+      )
+        merged[field] = previous[field] + detail[field];
+    }
+    result[position] = merged as (typeof result)[number];
+  }
+  return result;
+}

package/src/gateway/types.ts

@@ -45,6 +45,8 @@ export interface ChatMessage {
   role: ChatRole;
   content: string | readonly ChatContentPart[];
   name?: string;
+  reasoning?: string;
+  reasoningDetails?: readonly JsonValue[];
   toolCallId?: string;
   toolCalls?: readonly FunctionToolCall[];
 }
@@ -106,6 +108,8 @@ interface ChatCreateParamsBase {
   user?: string;
   /** Optional provider override. Omit to let the gateway select a provider. */
   provider?: string;
+  /** Options scoped to an upstream provider slug; other targets never receive them. */
+  providerOptions?: Readonly<Record<string, Readonly<Record<string, JsonValue>>>>;
   fallbackModels?: readonly string[];
   routingStrategy?: RoutingStrategy;
   metadata?: Readonly<Record<string, JsonValue>>;
@@ -167,6 +171,8 @@ export interface EmbeddingCreateParams {
   user?: string;
   /** Optional provider override. Omit to use the model's configured default provider. */
   provider?: string;
+  /** Options scoped to an upstream provider slug; other targets never receive them. */
+  providerOptions?: Readonly<Record<string, Readonly<Record<string, JsonValue>>>>;
   /** @deprecated Embeddings never fall back to another vector space. Requests with this option are rejected. */
   fallbackModels?: readonly string[];
   /** @deprecated Embeddings require direct model selection. */
@@ -204,6 +210,8 @@ export interface ImageGenerateParams {
   user?: string;
   /** Optional provider override. Omit to use the model's configured default provider. */
   provider?: string;
+  /** Options scoped to an upstream provider slug; other targets never receive them. */
+  providerOptions?: Readonly<Record<string, Readonly<Record<string, JsonValue>>>>;
   metadata?: Readonly<Record<string, JsonValue>>;
 }
 

package/src/index.ts

@@ -4,6 +4,10 @@ export type {
   NirmosAuth,
   EndpointExecution,
   KnowledgeMatch,
+  KnowledgeSearchRequest,
+  KnowledgeSearchResponse,
+  KnowledgeSearchMode,
+  KnowledgeSearchStatus,
   PromptTestRequest,
   PromptTestResponse,
 } from "./client.js";
@@ -137,3 +141,12 @@ export type {
 } from "./memory/types.js";
 
 export { managementManifest } from "./management/manifest.js";
+
+export {
+  extractPromptVariableNames,
+  interpolatePromptVariables,
+  promptVariableValue,
+  validatePromptTemplate,
+  validatePromptVariableValues,
+  validatePromptSettings,
+} from "./prompts/variables.js";