@smythos/sre 1.6.0 → 1.6.8

package/src/subsystems/LLMManager/LLM.service/connectors/Ollama.class.ts

@@ -0,0 +1,362 @@
+import { Ollama, ChatResponse } from 'ollama';
+import EventEmitter from 'events';
+
+import { JSON_RESPONSE_INSTRUCTION, BUILT_IN_MODEL_PREFIX } from '@sre/constants';
+import {
+    TLLMMessageBlock,
+    ToolData,
+    TLLMMessageRole,
+    APIKeySource,
+    TLLMEvent,
+    ILLMRequestFuncParams,
+    TLLMChatResponse,
+    ILLMRequestContext,
+    TLLMPreparedParams,
+    TLLMToolResultMessageBlock,
+    TLLMRequestBody,
+} from '@sre/types/LLM.types';
+import { LLMHelper } from '@sre/LLMManager/LLM.helper';
+
+import { LLMConnector } from '../LLMConnector';
+import { SystemEvents } from '@sre/Core/SystemEvents';
+import { Logger } from '@sre/helpers/Log.helper';
+
+const logger = Logger('OllamaConnector');
+
+type OllamaChatRequest = {
+    model: string;
+    messages: any[];
+    stream?: boolean;
+    options?: {
+        num_predict?: number;
+        temperature?: number;
+        top_p?: number;
+        top_k?: number;
+        stop?: string[];
+    };
+    tools?: any[];
+};
+
+export class OllamaConnector extends LLMConnector {
+    public name = 'LLM:Ollama';
+
+    private getClient(context: ILLMRequestContext): Ollama {
+        // Extract baseURL and sanitize it for Ollama SDK
+        let host = 'http://localhost:11434';
+
+        if (context.modelInfo.baseURL) {
+            // Handle baseURL that might include /api/ suffix
+            const baseURL = context.modelInfo.baseURL;
+            if (baseURL.endsWith('/api/')) {
+                // Remove /api/ suffix to get the root host
+                host = baseURL.replace(/\/api\/$/, '');
+            } else if (baseURL.endsWith('/api')) {
+                // Remove /api suffix
+                host = baseURL.replace(/\/api$/, '');
+            } else {
+                host = baseURL;
+            }
+        }
+
+        // No API key validation required for Ollama (local by default)
+        return new Ollama({ host });
+    }
+
+    protected async request({ acRequest, body, context }: ILLMRequestFuncParams): Promise<TLLMChatResponse> {
+        try {
+            logger.debug(`request ${this.name}`, acRequest.candidate);
+            const ollama = this.getClient(context);
+
+            const result = await ollama.chat({
+                ...body,
+                stream: false,
+            }) as unknown as ChatResponse;
+
+            const message = result.message;
+            const finishReason = result.done_reason || 'stop';
+            const usage = {
+                prompt_tokens: result.prompt_eval_count || 0,
+                completion_tokens: result.eval_count || 0,
+                total_tokens: (result.prompt_eval_count || 0) + (result.eval_count || 0),
+            };
+
+            this.reportUsage(usage, {
+                modelEntryName: context.modelEntryName,
+                keySource: context.isUserKey ? APIKeySource.User : APIKeySource.Smyth,
+                agentId: context.agentId,
+                teamId: context.teamId,
+            });
+
+            let toolsData: ToolData[] = [];
+            let useTool = false;
+
+            // Handle tool calls if present
+            if (message?.tool_calls) {
+                toolsData = message.tool_calls.map((tool, index) => ({
+                    index,
+                    id: tool.function?.name || `tool_${index}`,
+                    type: 'function',
+                    name: tool.function.name,
+                    arguments: tool.function.arguments,
+                    role: TLLMMessageRole.Assistant,
+                }));
+                useTool = true;
+            }
+
+            return {
+                content: message?.content ?? '',
+                finishReason,
+                useTool,
+                toolsData,
+                message: message as any,
+                usage,
+            };
+        } catch (error) {
+            logger.error(`request ${this.name}`, error, acRequest.candidate);
+            throw error;
+        }
+    }
+
+    protected async streamRequest({ acRequest, body, context }: ILLMRequestFuncParams): Promise<EventEmitter> {
+        try {
+            logger.debug(`streamRequest ${this.name}`, acRequest.candidate);
+            const emitter = new EventEmitter();
+            const usage_data = [];
+
+            const ollama = this.getClient(context);
+            const stream = await ollama.chat({
+                ...body,
+                stream: true,
+            }) as AsyncIterable<ChatResponse>;
+
+            let toolsData: ToolData[] = [];
+            let fullContent = '';
+
+            (async () => {
+                for await (const chunk of stream) {
+                    // Emit content deltas
+                    if (chunk.message?.content) {
+                        const content = chunk.message.content;
+                        fullContent += content;
+                        emitter.emit('content', content);
+                    }
+
+                    // Handle tool calls accumulation
+                    if (chunk.message?.tool_calls) {
+                        chunk.message.tool_calls.forEach((toolCall, index) => {
+                            if (!toolsData[index]) {
+                                toolsData[index] = {
+                                    index,
+                                    id: toolCall.function?.name || `tool_${index}`,
+                                    type: 'function',
+                                    name: toolCall.function?.name,
+                                    arguments: toolCall.function?.arguments || '',
+                                    role: 'assistant',
+                                };
+                            } else {
+                                // Merge arguments across chunks for string arguments
+                                if (typeof toolsData[index].arguments === 'string' && typeof toolCall.function?.arguments === 'string') {
+                                    toolsData[index].arguments += toolCall.function.arguments;
+                                } else {
+                                    // For object arguments, merge them properly
+                                    toolsData[index].arguments = { ...toolsData[index].arguments as any, ...toolCall.function?.arguments };
+                                }
+                            }
+                        });
+                    }
+
+                    // Capture usage data when available
+                    if (chunk.prompt_eval_count !== undefined || chunk.eval_count !== undefined) {
+                        const usage = {
+                            prompt_tokens: chunk.prompt_eval_count || 0,
+                            completion_tokens: chunk.eval_count || 0,
+                            total_tokens: (chunk.prompt_eval_count || 0) + (chunk.eval_count || 0),
+                        };
+                        usage_data.push(usage);
+                    }
+                }
+
+                // Emit tool info if tools were requested
+                if (toolsData.length > 0) {
+                    emitter.emit(TLLMEvent.ToolInfo, toolsData);
+                }
+
+                // Report usage
+                usage_data.forEach((usage) => {
+                    this.reportUsage(usage, {
+                        modelEntryName: context.modelEntryName,
+                        keySource: context.isUserKey ? APIKeySource.User : APIKeySource.Smyth,
+                        agentId: context.agentId,
+                        teamId: context.teamId,
+                    });
+                });
+
+                // Final end event
+                setTimeout(() => {
+                    emitter.emit('end', toolsData);
+                }, 100);
+            })();
+
+            return emitter;
+        } catch (error: any) {
+            logger.error(`streamRequest ${this.name}`, error, acRequest.candidate);
+            throw error;
+        }
+    }
+
+    protected async reqBodyAdapter(params: TLLMPreparedParams): Promise<TLLMRequestBody> {
+        const messages = params?.messages || [];
+
+        const body: OllamaChatRequest = {
+            model: params.model as string,
+            messages,
+        };
+
+        // Handle JSON response format
+        const responseFormat = params?.responseFormat || '';
+        if (responseFormat === 'json') {
+            if (messages?.[0]?.role === 'system') {
+                messages[0].content += JSON_RESPONSE_INSTRUCTION;
+            } else {
+                messages.unshift({ role: 'system', content: JSON_RESPONSE_INSTRUCTION });
+            }
+        }
+
+        // Map SRE options to Ollama options
+        const options: any = {};
+
+        if (params.maxTokens !== undefined) options.num_predict = params.maxTokens;
+        if (params.temperature !== undefined) options.temperature = params.temperature;
+        if (params.topP !== undefined) options.top_p = params.topP;
+        if (params.topK !== undefined) options.top_k = params.topK;
+        if (params.stopSequences?.length) options.stop = params.stopSequences;
+
+        if (Object.keys(options).length > 0) {
+            body.options = options;
+        }
+
+        // Handle tools
+        if (params.toolsConfig?.tools) {
+            body.tools = params.toolsConfig.tools.map(tool => ({
+                type: 'function',
+                function: {
+                    name: tool.function.name,
+                    description: tool.function.description,
+                    parameters: tool.function.parameters,
+                },
+            }));
+        }
+
+        return body as unknown as TLLMRequestBody;
+    }
+
+    protected reportUsage(
+        usage: { prompt_tokens: number; completion_tokens: number; total_tokens: number },
+        metadata: { modelEntryName: string; keySource: APIKeySource; agentId: string; teamId: string }
+    ) {
+        // SmythOS (built-in) models have a prefix, so we need to remove it to get the model name
+        const modelName = metadata.modelEntryName.replace(BUILT_IN_MODEL_PREFIX, '');
+
+        const usageData = {
+            sourceId: `llm:${modelName}`,
+            input_tokens: usage.prompt_tokens,
+            output_tokens: usage.completion_tokens,
+            input_tokens_cache_write: 0,
+            input_tokens_cache_read: 0,
+            keySource: metadata.keySource,
+            agentId: metadata.agentId,
+            teamId: metadata.teamId,
+        };
+        SystemEvents.emit('USAGE:LLM', usageData);
+
+        return usageData;
+    }
+
+    public transformToolMessageBlocks({
+        messageBlock,
+        toolsData,
+    }: {
+        messageBlock: TLLMMessageBlock;
+        toolsData: ToolData[];
+    }): TLLMToolResultMessageBlock[] {
+        const messageBlocks: TLLMToolResultMessageBlock[] = [];
+
+        // Transform the assistant message block if present
+        if (messageBlock) {
+            const transformedMessageBlock = {
+                ...messageBlock,
+                content: typeof messageBlock.content === 'object' ? JSON.stringify(messageBlock.content) : messageBlock.content,
+            };
+            if (transformedMessageBlock.tool_calls) {
+                for (let toolCall of transformedMessageBlock.tool_calls) {
+                    const args = toolCall?.function?.arguments;
+                    if (typeof args === 'string') {
+                        try {
+                            toolCall.function.arguments = JSON.parse(args);
+                        } catch {
+                            toolCall.function.arguments = {};
+                        }
+                    }
+                    // If it's already an object, keep as-is for Ollama
+                }
+            }
+            messageBlocks.push(transformedMessageBlock);
+        }
+
+        // Transform tool results into tool role messages
+        const transformedToolsData = toolsData.map((toolData) => ({
+            tool_call_id: toolData.id,
+            role: TLLMMessageRole.Tool,
+            name: toolData.name,
+            content: typeof toolData.result === 'string' ? toolData.result : JSON.stringify(toolData.result),
+        }));
+
+        return [...messageBlocks, ...transformedToolsData];
+    }
+
+    public formatToolsConfig({ type = 'function', toolDefinitions, toolChoice = 'auto' }) {
+        let tools = [];
+
+        if (type === 'function') {
+            tools = toolDefinitions.map((tool) => {
+                const { name, description, properties, requiredFields } = tool;
+
+                return {
+                    type: 'function',
+                    function: {
+                        name,
+                        description,
+                        parameters: {
+                            type: 'object',
+                            properties,
+                            required: requiredFields,
+                        },
+                    },
+                };
+            });
+        }
+
+        return tools?.length > 0 ? { tools, tool_choice: toolChoice } : {};
+    }
+
+    public getConsistentMessages(messages: TLLMMessageBlock[]): TLLMMessageBlock[] {
+        const _messages = LLMHelper.removeDuplicateUserMessages(messages);
+
+        return _messages.map((message) => {
+            const _message = { ...message };
+            let textContent = '';
+
+            if (message?.parts) {
+                textContent = message.parts.map((textBlock) => textBlock?.text || '').join(' ');
+            } else if (Array.isArray(message?.content)) {
+                textContent = message.content.map((textBlock) => textBlock?.text || '').join(' ');
+            } else if (message?.content) {
+                textContent = message.content as string;
+            }
+
+            _message.content = textContent;
+
+            return _message;
+        });
+    }
+}