smoltalk 0.0.34 → 0.0.35

package/dist/classes/message/AssistantMessage.d.ts

@@ -1,5 +1,5 @@
 import { BaseMessage, MessageClass } from "./BaseMessage.js";
-import { TextPart } from "../../types.js";
+import { TextPart, ThinkingBlock } from "../../types.js";
 import { ChatCompletionMessageParam } from "openai/resources";
 import { Content } from "@google/genai";
 import { ToolCall, ToolCallJSON } from "../ToolCall.js";
@@ -12,6 +12,7 @@ export type AssistantMessageJSON = {
     audio: any | null | undefined;
     refusal: string | null | undefined;
     toolCalls: ToolCallJSON[] | undefined;
+    thinkingBlocks: ThinkingBlock[] | undefined;
 };
 export declare class AssistantMessage extends BaseMessage implements MessageClass {
     _role: "assistant";
@@ -20,12 +21,14 @@ export declare class AssistantMessage extends BaseMessage implements MessageClas
     _audio?: any | null;
     _refusal?: string | null;
     _toolCalls?: ToolCall[];
+    _thinkingBlocks?: ThinkingBlock[];
     _rawData?: any;
     constructor(content: string | Array<TextPart> | null, options?: {
         name?: string;
         audio?: any | null;
         refusal?: string | null;
         toolCalls?: ToolCall[];
+        thinkingBlocks?: ThinkingBlock[];
         rawData?: any;
     });
     get content(): string;
@@ -36,6 +39,7 @@ export declare class AssistantMessage extends BaseMessage implements MessageClas
     get refusal(): string | null | undefined;
     get toolCalls(): ToolCall[] | undefined;
     get rawData(): any;
+    get thinkingBlocks(): ThinkingBlock[] | undefined;
     toJSON(): AssistantMessageJSON;
     static fromJSON(json: any): AssistantMessage;
     toOpenAIMessage(): ChatCompletionMessageParam;
@@ -45,6 +49,10 @@ export declare class AssistantMessage extends BaseMessage implements MessageClas
     toAnthropicMessage(): {
         role: "assistant";
         content: string | Array<{
+            type: "thinking";
+            thinking: string;
+            signature: string;
+        } | {
             type: "text";
             text: string;
         } | {

package/dist/classes/message/AssistantMessage.js

@@ -7,6 +7,7 @@ export class AssistantMessage extends BaseMessage {
     _audio;
     _refusal;
     _toolCalls;
+    _thinkingBlocks;
     _rawData;
     constructor(content, options = {}) {
         super();
@@ -15,6 +16,7 @@ export class AssistantMessage extends BaseMessage {
         this._audio = options.audio;
         this._refusal = options.refusal;
         this._toolCalls = options.toolCalls;
+        this._thinkingBlocks = options.thinkingBlocks;
         this._rawData = options.rawData;
     }
     get content() {
@@ -46,6 +48,9 @@ export class AssistantMessage extends BaseMessage {
     get rawData() {
         return this._rawData;
     }
+    get thinkingBlocks() {
+        return this._thinkingBlocks;
+    }
     toJSON() {
         return {
             role: this.role,
@@ -54,6 +59,7 @@ export class AssistantMessage extends BaseMessage {
             audio: this.audio,
             refusal: this.refusal,
             toolCalls: this.toolCalls?.map((tc) => tc.toJSON()),
+            thinkingBlocks: this._thinkingBlocks,
         };
     }
     static fromJSON(json) {
@@ -64,6 +70,7 @@ export class AssistantMessage extends BaseMessage {
             toolCalls: json.toolCalls
                 ? json.toolCalls.map((tcJson) => ToolCall.fromJSON(tcJson))
                 : undefined,
+            thinkingBlocks: json.thinkingBlocks,
             rawData: json.rawData,
         });
     }
@@ -95,6 +102,12 @@ export class AssistantMessage extends BaseMessage {
     }
     toGoogleMessage() {
         const parts = [];
+        // Prepend thought parts with their signatures so Gemini can resume reasoning
+        if (this._thinkingBlocks) {
+            for (const block of this._thinkingBlocks) {
+                parts.push({ thought: true, text: block.text, thoughtSignature: block.signature });
+            }
+        }
         if (this.content) {
             parts.push({ text: this.content });
         }
@@ -119,10 +132,18 @@ export class AssistantMessage extends BaseMessage {
                 ? this._content.length > 0
                 : this._content.length > 0);
         const hasToolCalls = this._toolCalls && this._toolCalls.length > 0;
-        if (!hasToolCalls) {
+        const hasThinking = this._thinkingBlocks && this._thinkingBlocks.length > 0;
+        // If only text and no thinking/tool calls, use string shorthand
+        if (!hasToolCalls && !hasThinking) {
             return { role: "assistant", content: this.content };
         }
         const blocks = [];
+        // Thinking blocks must come first (Anthropic requires this ordering)
+        if (hasThinking) {
+            for (const block of this._thinkingBlocks) {
+                blocks.push({ type: "thinking", thinking: block.text, signature: block.signature });
+            }
+        }
         if (hasText) {
             const text = typeof this._content === "string"
                 ? this._content
@@ -131,7 +152,7 @@ export class AssistantMessage extends BaseMessage {
                 blocks.push({ type: "text", text });
             }
         }
-        for (const tc of this._toolCalls) {
+        for (const tc of this._toolCalls ?? []) {
             blocks.push({ type: "tool_use", id: tc.id, name: tc.name, input: tc.arguments });
         }
         return { role: "assistant", content: blocks };

package/dist/classes/message/index.d.ts

@@ -25,6 +25,10 @@ export declare function assistantMessage(content: string | Array<TextPart> | nul
     audio?: any | null;
     refusal?: string | null;
     toolCalls?: Array<any>;
+    thinkingBlocks?: Array<{
+        text: string;
+        signature: string;
+    }>;
     rawData?: any;
 }): AssistantMessage;
 export declare function developerMessage(content: string | Array<TextPart>, options?: {

package/dist/clients/anthropic.js

@@ -64,16 +64,20 @@ export class SmolAnthropic extends BaseClient {
                 description: tool.description,
             }))
             : undefined;
-        return { system, messages: anthropicMessages, tools };
+        const thinking = config.thinking?.enabled
+            ? { type: "enabled", budget_tokens: config.thinking.budgetTokens ?? 5000 }
+            : undefined;
+        return { system, messages: anthropicMessages, tools, thinking };
     }
     async _textSync(config) {
-        const { system, messages, tools } = this.buildRequest(config);
+        const { system, messages, tools, thinking } = this.buildRequest(config);
         this.logger.debug("Sending request to Anthropic:", {
             model: this.model,
             max_tokens: config.maxTokens ?? DEFAULT_MAX_TOKENS,
             messages,
             system,
             tools,
+            thinking,
         });
         const response = await this.client.messages.create({
             model: this.model,
@@ -81,6 +85,7 @@ export class SmolAnthropic extends BaseClient {
             messages,
             ...(system && { system }),
             ...(tools && { tools }),
+            ...(thinking && { thinking }),
             ...(config.temperature !== undefined && {
                 temperature: config.temperature,
             }),
@@ -90,6 +95,7 @@ export class SmolAnthropic extends BaseClient {
         this.logger.debug("Response from Anthropic:", response);
         let output = null;
         const toolCalls = [];
+        const thinkingBlocks = [];
         for (const block of response.content) {
             if (block.type === "text") {
                 output = (output ?? "") + block.text;
@@ -97,24 +103,30 @@ export class SmolAnthropic extends BaseClient {
             else if (block.type === "tool_use") {
                 toolCalls.push(new ToolCall(block.id, block.name, block.input));
             }
+            else if (block.type === "thinking") {
+                const b = block;
+                thinkingBlocks.push({ text: b.thinking, signature: b.signature });
+            }
         }
         const { usage, cost } = this.calculateUsageAndCost(response.usage);
         return success({
             output,
             toolCalls,
+            ...(thinkingBlocks.length > 0 && { thinkingBlocks }),
             usage,
             cost,
             model: this.model,
         });
     }
     async *_textStream(config) {
-        const { system, messages, tools } = this.buildRequest(config);
+        const { system, messages, tools, thinking } = this.buildRequest(config);
         this.logger.debug("Sending streaming request to Anthropic:", {
             model: this.model,
             max_tokens: config.maxTokens ?? DEFAULT_MAX_TOKENS,
             messages,
             system,
             tools,
+            thinking,
         });
         const stream = await this.client.messages.create({
             model: this.model,
@@ -122,6 +134,7 @@ export class SmolAnthropic extends BaseClient {
             messages,
             ...(system && { system }),
             ...(tools && { tools }),
+            ...(thinking && { thinking }),
             ...(config.temperature !== undefined && {
                 temperature: config.temperature,
             }),
@@ -131,19 +144,25 @@ export class SmolAnthropic extends BaseClient {
         let content = "";
         // Track tool blocks by index: index -> { id, name, arguments (partial JSON) }
         const toolBlocks = new Map();
+        // Track thinking blocks by index: index -> { text, signature }
+        const thinkingBlockMap = new Map();
         let inputTokens = 0;
         let outputTokens = 0;
         for await (const event of stream) {
             if (event.type === "message_start") {
                 inputTokens = event.message.usage.input_tokens;
             }
-            else if (event.type === "content_block_start" &&
-                event.content_block.type === "tool_use") {
-                toolBlocks.set(event.index, {
-                    id: event.content_block.id,
-                    name: event.content_block.name,
-                    arguments: "",
-                });
+            else if (event.type === "content_block_start") {
+                if (event.content_block.type === "tool_use") {
+                    toolBlocks.set(event.index, {
+                        id: event.content_block.id,
+                        name: event.content_block.name,
+                        arguments: "",
+                    });
+                }
+                else if (event.content_block.type === "thinking") {
+                    thinkingBlockMap.set(event.index, { text: "", signature: "" });
+                }
             }
             else if (event.type === "content_block_delta") {
                 if (event.delta.type === "text_delta") {
@@ -156,6 +175,25 @@ export class SmolAnthropic extends BaseClient {
                         block.arguments += event.delta.partial_json;
                     }
                 }
+                else if (event.delta.type === "thinking_delta") {
+                    const block = thinkingBlockMap.get(event.index);
+                    if (block) {
+                        block.text += event.delta.thinking;
+                    }
+                }
+                else if (event.delta.type === "signature_delta") {
+                    const block = thinkingBlockMap.get(event.index);
+                    if (block) {
+                        block.signature = event.delta.signature;
+                    }
+                }
+            }
+            else if (event.type === "content_block_stop") {
+                // Emit thinking chunk once the block is fully assembled
+                const thinkingBlock = thinkingBlockMap.get(event.index);
+                if (thinkingBlock) {
+                    yield { type: "thinking", text: thinkingBlock.text, signature: thinkingBlock.signature };
+                }
             }
             else if (event.type === "message_delta") {
                 outputTokens = event.usage.output_tokens;
@@ -168,6 +206,7 @@ export class SmolAnthropic extends BaseClient {
             toolCalls.push(toolCall);
             yield { type: "tool_call", toolCall };
         }
+        const thinkingBlocks = Array.from(thinkingBlockMap.values());
         const usage = {
             inputTokens,
             outputTokens,
@@ -179,6 +218,7 @@ export class SmolAnthropic extends BaseClient {
             result: {
                 output: content || null,
                 toolCalls,
+                ...(thinkingBlocks.length > 0 && { thinkingBlocks }),
                 usage,
                 cost,
                 model: this.model,

package/dist/clients/google.js

@@ -91,6 +91,7 @@ export class SmolGoogle extends BaseClient {
         this.logger.debug("Response from Google Gemini:", JSON.stringify(result, null, 2));
         const output = result.text || null;
         const toolCalls = [];
+        const thinkingBlocks = [];
         result.candidates?.forEach((candidate) => {
             if (candidate.content && candidate.content.parts) {
                 candidate.content.parts.forEach((part) => {
@@ -98,6 +99,13 @@ export class SmolGoogle extends BaseClient {
                         const functionCall = part.functionCall;
                         toolCalls.push(new ToolCall("", functionCall.name, functionCall.args));
                     }
+                    // Capture thought parts (thought: true indicates a thinking part)
+                    if (part.thoughtSignature) {
+                        thinkingBlocks.push({
+                            text: part.text || "",
+                            signature: part.thoughtSignature,
+                        });
+                    }
                 });
             }
         });
@@ -107,6 +115,7 @@ export class SmolGoogle extends BaseClient {
         return success({
             output,
             toolCalls,
+            ...(thinkingBlocks.length > 0 && { thinkingBlocks }),
             usage,
             cost,
             model: request.model,
@@ -118,6 +127,7 @@ export class SmolGoogle extends BaseClient {
         const stream = await this.client.models.generateContentStream(request);
         let content = "";
         const toolCallsMap = new Map();
+        const thinkingBlocks = [];
         let usage;
         let cost;
         for await (const chunk of stream) {
@@ -127,22 +137,28 @@ export class SmolGoogle extends BaseClient {
                 usage = usageAndCost.usage;
                 cost = usageAndCost.cost;
             }
-            // Handle text content
-            if (chunk.text) {
-                content += chunk.text;
-                yield { type: "text", text: chunk.text };
-            }
-            // Handle function calls
-            if (chunk.functionCalls) {
-                for (const functionCall of chunk.functionCalls) {
-                    const id = functionCall.id || functionCall.name || "";
-                    const name = functionCall.name || "";
-                    if (!toolCallsMap.has(id)) {
-                        toolCallsMap.set(id, {
-                            id: id,
-                            name: name,
-                            arguments: functionCall.args,
-                        });
+            // Iterate raw parts to capture thought signatures and regular content
+            for (const candidate of chunk.candidates || []) {
+                for (const part of candidate?.content?.parts || []) {
+                    const p = part;
+                    if (p.thoughtSignature) {
+                        const block = {
+                            text: p.text || "",
+                            signature: p.thoughtSignature,
+                        };
+                        thinkingBlocks.push(block);
+                        yield { type: "thinking", text: block.text, signature: block.signature };
+                    }
+                    else if (p.text) {
+                        content += p.text;
+                        yield { type: "text", text: p.text };
+                    }
+                    else if (p.functionCall) {
+                        const id = p.functionCall.id || p.functionCall.name || "";
+                        const name = p.functionCall.name || "";
+                        if (!toolCallsMap.has(id)) {
+                            toolCallsMap.set(id, { id, name, arguments: p.functionCall.args });
+                        }
                     }
                 }
             }
@@ -160,6 +176,7 @@ export class SmolGoogle extends BaseClient {
             result: {
                 output: content || null,
                 toolCalls,
+                ...(thinkingBlocks.length > 0 && { thinkingBlocks }),
                 usage,
                 cost,
                 model: request.model,

package/dist/types.d.ts

@@ -5,6 +5,10 @@ import { Message } from "./classes/message/index.js";
 import { ToolCall } from "./classes/ToolCall.js";
 import { ModelConfig, ModelName, Provider } from "./models.js";
 import { Result } from "./types/result.js";
+export type ThinkingBlock = {
+    text: string;
+    signature: string;
+};
 export type PromptConfig = {
     messages: Message[];
     tools?: {
@@ -19,6 +23,10 @@ export type PromptConfig = {
     parallelToolCalls?: boolean;
     responseFormat?: ZodType;
     stream?: boolean;
+    thinking?: {
+        enabled: boolean;
+        budgetTokens?: number;
+    };
     responseFormatOptions?: Partial<{
         name: string;
         strict: boolean;
@@ -65,6 +73,7 @@ export type CostEstimate = {
 export type PromptResult = {
     output: string | null;
     toolCalls: ToolCall[];
+    thinkingBlocks?: ThinkingBlock[];
     usage?: TokenUsage;
     cost?: CostEstimate;
     model?: ModelName | ModelConfig;
@@ -72,6 +81,10 @@ export type PromptResult = {
 export type StreamChunk = {
     type: "text";
     text: string;
+} | {
+    type: "thinking";
+    text: string;
+    signature?: string;
 } | {
     type: "tool_call";
     toolCall: ToolCall;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "smoltalk",
-  "version": "0.0.34",
+  "version": "0.0.35",
   "description": "A common interface for LLM APIs",
   "homepage": "https://github.com/egonSchiele/smoltalk",
   "scripts": {
@@ -47,4 +47,4 @@
     "ollama": "^0.6.3",
     "openai": "^6.15.0"
   }
-}
+}