@theokit/sdk 2.11.0 → 2.11.2

package/dist/internal/agent-loop/message-builders.d.ts

@@ -1,6 +1,8 @@
-import type { SDKAssistantMessage, SDKUserMessageEvent } from "../../types/messages.js";
+import type { SDKAssistantMessage, SDKThinkingMessage, SDKUserMessageEvent } from "../../types/messages.js";
 import type { LlmMessage, LlmToolCallPart } from "../llm/types.js";
 import type { AgentLoopInputs } from "./loop-types.js";
 export declare function buildUserEvent(inputs: AgentLoopInputs): SDKUserMessageEvent;
 export declare function buildAssistantEvent(inputs: AgentLoopInputs, text: string): SDKAssistantMessage;
+/** issue #47: the reasoning the model produced this turn, as a `thinking` SDKMessage. */
+export declare function buildThinkingEvent(inputs: AgentLoopInputs, text: string): SDKThinkingMessage;
 export declare function buildAssistantTurn(text: string, toolCalls: LlmToolCallPart[]): LlmMessage;

package/dist/internal/llm/openai.d.ts

@@ -1,4 +1,48 @@
 import type { LlmClient, LlmEvent, LlmFinish, LlmRequest } from "./types.js";
+interface OpenAIDeltaChunk {
+    choices?: Array<{
+        index: number;
+        delta?: {
+            content?: string;
+            /** OpenRouter unified reasoning delta (issue #47) — streamed separately from `content`. */
+            reasoning?: string;
+            /**
+             * Sibling reasoning field used by some OpenAI-compatible providers (DeepSeek's direct API,
+             * many vLLM / LMStudio reasoning parsers). Treated identically to `reasoning` (issue #47,
+             * F-domain-2). OpenRouter normalizes to `reasoning`; this is the fallback for non-OpenRouter
+             * compat endpoints reached via `baseUrl` override.
+             */
+            reasoning_content?: string;
+            tool_calls?: Array<{
+                index: number;
+                id?: string;
+                function?: {
+                    name?: string;
+                    arguments?: string;
+                };
+            }>;
+        };
+        finish_reason?: string | null;
+    }>;
+    usage?: {
+        prompt_tokens?: number;
+        completion_tokens?: number;
+        /**
+         * OpenAI + OpenRouter extended usage (D376). Anthropic prompt caching
+         * surfaces as `prompt_tokens_details.cached_tokens` on OpenRouter passthrough;
+         * native OpenAI exposes the same field on `chatcmpl-*` responses.
+         */
+        prompt_tokens_details?: {
+            cached_tokens?: number;
+        };
+        completion_tokens_details?: {
+            reasoning_tokens?: number;
+        };
+        /** Anthropic-on-OpenRouter passes cache_creation_tokens top-level (cline#10266). */
+        cache_creation_input_tokens?: number;
+        cache_read_input_tokens?: number;
+    };
+}
 export declare class OpenAIClient implements LlmClient {
     private readonly options;
     readonly name = "openai";
@@ -7,3 +51,23 @@ export declare class OpenAIClient implements LlmClient {
     constructor(options: OpenAIClientOptions);
     stream(request: LlmRequest, signal: AbortSignal): AsyncGenerator<LlmEvent, LlmFinish, void>;
 }
+declare class OpenAIStreamAccumulator {
+    private text;
+    private stopReason;
+    private inputTokens?;
+    private outputTokens?;
+    private cacheReadTokens?;
+    private cacheWriteTokens?;
+    private reasoningTokens?;
+    private readonly toolCalls;
+    consume(chunk: OpenAIDeltaChunk): LlmEvent[];
+    private applyReasoningDelta;
+    private applyUsage;
+    private applyContentDelta;
+    private mergeToolCallDeltas;
+    private applyFinishReason;
+    finish(): LlmFinish;
+}
+/** issue #47 test seam — exercise reasoning-delta parsing without a live stream. */
+export declare const __testing__OpenAIStreamAccumulator: typeof OpenAIStreamAccumulator;
+export {};

package/dist/internal/llm/types.d.ts

@@ -80,10 +80,33 @@ export interface LlmRequest {
     temperature?: number;
     /** T3.6 — opt into native structured outputs (OpenAI-compat providers). */
     responseFormat?: LlmResponseFormat;
+    /**
+     * Reasoning / extended-thinking request (issue #47). Derived from `ModelSelection.params` (the
+     * `thinking` param). The wire encoding is provider-specific: OpenRouter (and OpenAI-compatible
+     * passthroughs) use the unified `reasoning: { effort }` object; native OpenAI Chat Completions uses
+     * the top-level `reasoning_effort` string (see `buildOpenAIBody`). The model streams reasoning back
+     * as `delta.reasoning` (or `delta.reasoning_content` on some compat providers), surfaced as
+     * `reasoning_delta` events. `effort` is required once the object is present — an empty object would
+     * request nothing.
+     */
+    reasoning?: {
+        effort: string;
+    };
+    /**
+     * Step-cap force-close: per-request tool gating. `"none"` tells the provider to emit no tool
+     * calls even when `tools` are present (forcing a text answer); `"required"` forces a tool call;
+     * `"auto"` (or omitted) is the default. Maps to OpenAI/OpenRouter `tool_choice`. The agent loop
+     * sets `"none"` on a final/ceiling round so a cached agent (whose tools cannot be un-registered)
+     * is still forced to produce a closing summary.
+     */
+    toolChoice?: "auto" | "none" | "required";
 }
 export type LlmEvent = {
     type: "text_delta";
     text: string;
+} | {
+    type: "reasoning_delta";
+    text: string;
 } | {
     type: "tool_use";
     id: string;

package/dist/{run-D22b53SU.d.cts → run-TMdc7gmo.d.cts}

@@ -827,6 +827,13 @@ interface SendOptions {
     onDelta?: (args: {
         update: InteractionUpdate;
     }) => void | Promise<void>;
+    /**
+     * Per-call tool gate (OpenAI/OpenRouter `tool_choice`). `"none"` forces a text answer even when
+     * the agent has tools registered — used by an agent loop to force a closing summary at its step
+     * ceiling (a cached agent's tools cannot be un-registered, so the gate is per-send). `"required"`
+     * forces a tool call; `"auto"` (or omitted) is the default. Local runtime; OpenAI-compat providers.
+     */
+    toolChoice?: "auto" | "none" | "required";
     /** Local agents only. Expire a stuck active run before starting this message. */
     local?: {
         force?: boolean;

package/dist/{run-D22b53SU.d.ts → run-TMdc7gmo.d.ts}

@@ -827,6 +827,13 @@ interface SendOptions {
     onDelta?: (args: {
         update: InteractionUpdate;
     }) => void | Promise<void>;
+    /**
+     * Per-call tool gate (OpenAI/OpenRouter `tool_choice`). `"none"` forces a text answer even when
+     * the agent has tools registered — used by an agent loop to force a closing summary at its step
+     * ceiling (a cached agent's tools cannot be un-registered, so the gate is per-send). `"required"`
+     * forces a tool call; `"auto"` (or omitted) is the default. Local runtime; OpenAI-compat providers.
+     */
+    toolChoice?: "auto" | "none" | "required";
     /** Local agents only. Expire a stuck active run before starting this message. */
     local?: {
         force?: boolean;

package/dist/types/run.d.ts

@@ -221,6 +221,13 @@ export interface SendOptions {
     onDelta?: (args: {
         update: InteractionUpdate;
     }) => void | Promise<void>;
+    /**
+     * Per-call tool gate (OpenAI/OpenRouter `tool_choice`). `"none"` forces a text answer even when
+     * the agent has tools registered — used by an agent loop to force a closing summary at its step
+     * ceiling (a cached agent's tools cannot be un-registered, so the gate is per-send). `"required"`
+     * forces a tool call; `"auto"` (or omitted) is the default. Local runtime; OpenAI-compat providers.
+     */
+    toolChoice?: "auto" | "none" | "required";
     /** Local agents only. Expire a stuck active run before starting this message. */
     local?: {
         force?: boolean;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@theokit/sdk",
-  "version": "2.11.0",
+  "version": "2.11.2",
   "description": "TypeScript SDK for the Theo agent harness — same surface, local or cloud.",
   "license": "Apache-2.0",
   "homepage": "https://github.com/usetheo/theokit-sdk#readme",
@@ -298,15 +298,6 @@
     "**/agent.js",
     "**/agent.cjs"
   ],
-  "scripts": {
-    "build": "tsup && cp src/internal/providers/provider-catalog.json dist/provider-catalog.json",
-    "test": "vitest run --no-file-parallelism",
-    "test:watch": "vitest",
-    "typecheck": "tsc --noEmit",
-    "clean": "rm -rf dist",
-    "docs:json": "typedoc --options typedoc.json",
-    "docs:drift": "tsx scripts/check-docs-drift.ts"
-  },
   "peerDependencies": {
     "@lancedb/lancedb": "^0.30.0",
     "@types/ws": ">=8.0.0",
@@ -363,8 +354,6 @@
   "devDependencies": {
     "@opentelemetry/api": "^1.9.1",
     "@opentelemetry/sdk-trace-base": "^1.30.1",
-    "@theokit/sdk-handoff": "workspace:*",
-    "@theokit/sdk-memory": "workspace:*",
     "@types/better-sqlite3": "^7.6.13",
     "@types/proper-lockfile": "^4.1.4",
     "@types/ws": "^8.18.0",
@@ -374,6 +363,17 @@
     "sqlite-vec": "^0.1.9",
     "typedoc": "^0.28.19",
     "ws": "^8.18.0",
-    "zod": "^4.0.0"
+    "zod": "^4.0.0",
+    "@theokit/sdk-handoff": "0.1.0",
+    "@theokit/sdk-memory": "0.2.0"
+  },
+  "scripts": {
+    "build": "tsup && cp src/internal/providers/provider-catalog.json dist/provider-catalog.json",
+    "test": "vitest run --no-file-parallelism",
+    "test:watch": "vitest",
+    "typecheck": "tsc --noEmit",
+    "clean": "rm -rf dist",
+    "docs:json": "typedoc --options typedoc.json",
+    "docs:drift": "tsx scripts/check-docs-drift.ts"
   }
-}
+}