feat(providers): implement custom classes

src/lib/models/providers/ollama/index.ts

@@ -0,0 +1,136 @@
+import { UIConfigField } from '@/lib/config/types';
+import { getConfiguredModelProviderById } from '@/lib/config/serverRegistry';
+import BaseModelProvider from '../../base/provider';
+import { Model, ModelList, ProviderMetadata } from '../../types';
+import BaseLLM from '../../base/llm';
+import BaseEmbedding from '../../base/embedding';
+import OllamaLLM from './ollamaLLM';
+import OllamaEmbedding from './ollamaEmbedding';
+
+interface OllamaConfig {
+  baseURL: string;
+}
+
+const providerConfigFields: UIConfigField[] = [
+  {
+    type: 'string',
+    name: 'Base URL',
+    key: 'baseURL',
+    description: 'The base URL for the Ollama',
+    required: true,
+    placeholder: process.env.DOCKER
+      ? 'http://host.docker.internal:11434'
+      : 'http://localhost:11434',
+    env: 'OLLAMA_BASE_URL',
+    scope: 'server',
+  },
+];
+
+class OllamaProvider extends BaseModelProvider<OllamaConfig> {
+  constructor(id: string, name: string, config: OllamaConfig) {
+    super(id, name, config);
+  }
+
+  async getDefaultModels(): Promise<ModelList> {
+    try {
+      const res = await fetch(`${this.config.baseURL}/api/tags`, {
+        method: 'GET',
+        headers: {
+          'Content-type': 'application/json',
+        },
+      });
+
+      const data = await res.json();
+
+      const models: Model[] = data.models.map((m: any) => {
+        return {
+          name: m.name,
+          key: m.model,
+        };
+      });
+
+      return {
+        embedding: models,
+        chat: models,
+      };
+    } catch (err) {
+      if (err instanceof TypeError) {
+        throw new Error(
+          'Error connecting to Ollama API. Please ensure the base URL is correct and the Ollama server is running.',
+        );
+      }
+
+      throw err;
+    }
+  }
+
+  async getModelList(): Promise<ModelList> {
+    const defaultModels = await this.getDefaultModels();
+    const configProvider = getConfiguredModelProviderById(this.id)!;
+
+    return {
+      embedding: [
+        ...defaultModels.embedding,
+        ...configProvider.embeddingModels,
+      ],
+      chat: [...defaultModels.chat, ...configProvider.chatModels],
+    };
+  }
+
+  async loadChatModel(key: string): Promise<BaseLLM<any>> {
+    const modelList = await this.getModelList();
+
+    const exists = modelList.chat.find((m) => m.key === key);
+
+    if (!exists) {
+      throw new Error(
+        'Error Loading Ollama Chat Model. Invalid Model Selected',
+      );
+    }
+
+    return new OllamaLLM({
+      baseURL: this.config.baseURL,
+      model: key,
+    });
+  }
+
+  async loadEmbeddingModel(key: string): Promise<BaseEmbedding<any>> {
+    const modelList = await this.getModelList();
+    const exists = modelList.embedding.find((m) => m.key === key);
+
+    if (!exists) {
+      throw new Error(
+        'Error Loading Ollama Embedding Model. Invalid Model Selected.',
+      );
+    }
+
+    return new OllamaEmbedding({
+      model: key,
+      baseURL: this.config.baseURL,
+    });
+  }
+
+  static parseAndValidate(raw: any): OllamaConfig {
+    if (!raw || typeof raw !== 'object')
+      throw new Error('Invalid config provided. Expected object');
+    if (!raw.baseURL)
+      throw new Error('Invalid config provided. Base URL must be provided');
+
+    return {
+      baseURL: String(raw.baseURL),
+    };
+  }
+
+  static getProviderConfigFields(): UIConfigField[] {
+    return providerConfigFields;
+  }
+
+  static getProviderMetadata(): ProviderMetadata {
+    return {
+      key: 'ollama',
+      name: 'Ollama',
+    };
+  }
+}
+
+export default OllamaProvider;

src/lib/models/providers/ollama/ollamaEmbedding.ts

@@ -0,0 +1,39 @@
+import { Ollama } from 'ollama';
+import BaseEmbedding from '../../base/embedding';
+
+type OllamaConfig = {
+  model: string;
+  baseURL?: string;
+};
+
+class OllamaEmbedding extends BaseEmbedding<OllamaConfig> {
+  ollamaClient: Ollama;
+
+  constructor(protected config: OllamaConfig) {
+    super(config);
+
+    this.ollamaClient = new Ollama({
+      host: this.config.baseURL || 'http://localhost:11434',
+    });
+  }
+
+  async embedText(texts: string[]): Promise<number[][]> {
+    const response = await this.ollamaClient.embed({
+      input: texts,
+      model: this.config.model,
+    });
+
+    return response.embeddings;
+  }
+
+  async embedChunks(chunks: Chunk[]): Promise<number[][]> {
+    const response = await this.ollamaClient.embed({
+      input: chunks.map((c) => c.content),
+      model: this.config.model,
+    });
+
+    return response.embeddings;
+  }
+}
+
+export default OllamaEmbedding;

src/lib/models/providers/ollama/ollamaLLM.ts

@@ -0,0 +1,149 @@
+import z from 'zod';
+import BaseLLM from '../../base/llm';
+import {
+  GenerateObjectInput,
+  GenerateOptions,
+  GenerateTextInput,
+  GenerateTextOutput,
+  StreamTextOutput,
+} from '../../types';
+import { Ollama } from 'ollama';
+import { parse } from 'partial-json';
+
+type OllamaConfig = {
+  baseURL: string;
+  model: string;
+  options?: GenerateOptions;
+};
+
+class OllamaLLM extends BaseLLM<OllamaConfig> {
+  ollamaClient: Ollama;
+
+  constructor(protected config: OllamaConfig) {
+    super(config);
+
+    this.ollamaClient = new Ollama({
+      host: this.config.baseURL || 'http://localhost:11434',
+    });
+  }
+
+  withOptions(options: GenerateOptions) {
+    this.config.options = {
+      ...this.config.options,
+      ...options,
+    };
+    return this;
+  }
+
+  async generateText(input: GenerateTextInput): Promise<GenerateTextOutput> {
+    this.withOptions(input.options || {});
+
+    const res = await this.ollamaClient.chat({
+      model: this.config.model,
+      messages: input.messages,
+      options: {
+        top_p: this.config.options?.topP,
+        temperature: this.config.options?.temperature,
+        num_predict: this.config.options?.maxTokens,
+        frequency_penalty: this.config.options?.frequencyPenalty,
+        presence_penalty: this.config.options?.presencePenalty,
+        stop: this.config.options?.stopSequences,
+      },
+    });
+
+    return {
+      content: res.message.content,
+      additionalInfo: {
+        reasoning: res.message.thinking,
+      },
+    };
+  }
+
+  async *streamText(
+    input: GenerateTextInput,
+  ): AsyncGenerator<StreamTextOutput> {
+    this.withOptions(input.options || {});
+
+    const stream = await this.ollamaClient.chat({
+      model: this.config.model,
+      messages: input.messages,
+      stream: true,
+      options: {
+        top_p: this.config.options?.topP,
+        temperature: this.config.options?.temperature,
+        num_predict: this.config.options?.maxTokens,
+        frequency_penalty: this.config.options?.frequencyPenalty,
+        presence_penalty: this.config.options?.presencePenalty,
+        stop: this.config.options?.stopSequences,
+      },
+    });
+
+    for await (const chunk of stream) {
+      yield {
+        contentChunk: chunk.message.content,
+        done: chunk.done,
+        additionalInfo: {
+          reasoning: chunk.message.thinking,
+        },
+      };
+    }
+  }
+
+  async generateObject<T>(input: GenerateObjectInput): Promise<T> {
+    this.withOptions(input.options || {});
+
+    const response = await this.ollamaClient.chat({
+      model: this.config.model,
+      messages: input.messages,
+      format: z.toJSONSchema(input.schema),
+      options: {
+        top_p: this.config.options?.topP,
+        temperature: this.config.options?.temperature,
+        num_predict: this.config.options?.maxTokens,
+        frequency_penalty: this.config.options?.frequencyPenalty,
+        presence_penalty: this.config.options?.presencePenalty,
+        stop: this.config.options?.stopSequences,
+      },
+    });
+
+    try {
+      return input.schema.parse(JSON.parse(response.message.content)) as T;
+    } catch (err) {
+      throw new Error(`Error parsing response from Ollama: ${err}`);
+    }
+  }
+
+  async *streamObject<T>(input: GenerateObjectInput): AsyncGenerator<T> {
+    let recievedObj: string = '';
+
+    this.withOptions(input.options || {});
+
+    const stream = await this.ollamaClient.chat({
+      model: this.config.model,
+      messages: input.messages,
+      format: z.toJSONSchema(input.schema),
+      stream: true,
+      options: {
+        top_p: this.config.options?.topP,
+        temperature: this.config.options?.temperature,
+        num_predict: this.config.options?.maxTokens,
+        frequency_penalty: this.config.options?.frequencyPenalty,
+        presence_penalty: this.config.options?.presencePenalty,
+        stop: this.config.options?.stopSequences,
+      },
+    });
+
+    for await (const chunk of stream) {
+      recievedObj += chunk.message.content;
+
+      try {
+        yield parse(recievedObj) as T;
+      } catch (err) {
+        console.log('Error parsing partial object from Ollama:', err);
+        yield {} as T;
+      }
+    }
+  }
+}
+
+export default OllamaLLM;