feat: монорепо миграция, Discover/SearxNG улучшения

- Миграция на монорепозиторий (apps/frontend, apps/chat-service, etc.)
- Discover: проверка SearxNG, понятное empty state при ненастроенном поиске
- searxng.ts: валидация URL, проверка JSON-ответа, авто-добавление http://
- docker/searxng-config: настройки для JSON API SearxNG

Co-authored-by: Cursor <cursoragent@cursor.com>

apps/llm-proxy/src/lib/models/base/embedding.ts

@@ -0,0 +1,9 @@
+import { Chunk } from '@/lib/types';
+
+abstract class BaseEmbedding<CONFIG> {
+  constructor(protected config: CONFIG) {}
+  abstract embedText(texts: string[]): Promise<number[][]>;
+  abstract embedChunks(chunks: Chunk[]): Promise<number[][]>;
+}
+
+export default BaseEmbedding;

apps/llm-proxy/src/lib/models/base/llm.ts

@@ -0,0 +1,22 @@
+import z from 'zod';
+import {
+  GenerateObjectInput,
+  GenerateOptions,
+  GenerateTextInput,
+  GenerateTextOutput,
+  StreamTextOutput,
+} from '../types';
+
+abstract class BaseLLM<CONFIG> {
+  constructor(protected config: CONFIG) {}
+  abstract generateText(input: GenerateTextInput): Promise<GenerateTextOutput>;
+  abstract streamText(
+    input: GenerateTextInput,
+  ): AsyncGenerator<StreamTextOutput>;
+  abstract generateObject<T>(input: GenerateObjectInput): Promise<z.infer<T>>;
+  abstract streamObject<T>(
+    input: GenerateObjectInput,
+  ): AsyncGenerator<Partial<z.infer<T>>>;
+}
+
+export default BaseLLM;

apps/llm-proxy/src/lib/models/base/provider.ts

@@ -0,0 +1,45 @@
+import { ModelList, ProviderMetadata } from '../types';
+import { UIConfigField } from '@/lib/config/types';
+import BaseLLM from './llm';
+import BaseEmbedding from './embedding';
+
+abstract class BaseModelProvider<CONFIG> {
+  constructor(
+    protected id: string,
+    protected name: string,
+    protected config: CONFIG,
+  ) {}
+  abstract getDefaultModels(): Promise<ModelList>;
+  abstract getModelList(): Promise<ModelList>;
+  abstract loadChatModel(modelName: string): Promise<BaseLLM<any>>;
+  abstract loadEmbeddingModel(modelName: string): Promise<BaseEmbedding<any>>;
+  static getProviderConfigFields(): UIConfigField[] {
+    throw new Error('Method not implemented.');
+  }
+  static getProviderMetadata(): ProviderMetadata {
+    throw new Error('Method not Implemented.');
+  }
+  static parseAndValidate(raw: any): any {
+    /* Static methods can't access class type parameters */
+    throw new Error('Method not Implemented.');
+  }
+}
+
+export type ProviderConstructor<CONFIG> = {
+  new (id: string, name: string, config: CONFIG): BaseModelProvider<CONFIG>;
+  parseAndValidate(raw: any): CONFIG;
+  getProviderConfigFields: () => UIConfigField[];
+  getProviderMetadata: () => ProviderMetadata;
+};
+
+export const createProviderInstance = <P extends ProviderConstructor<any>>(
+  Provider: P,
+  id: string,
+  name: string,
+  rawConfig: unknown,
+): InstanceType<P> => {
+  const cfg = Provider.parseAndValidate(rawConfig);
+  return new Provider(id, name, cfg) as InstanceType<P>;
+};
+
+export default BaseModelProvider;

apps/llm-proxy/src/lib/models/providers/anthropic/anthropicLLM.ts

@@ -0,0 +1,5 @@
+import OpenAILLM from '../openai/openaiLLM';
+
+class AnthropicLLM extends OpenAILLM {}
+
+export default AnthropicLLM;

apps/llm-proxy/src/lib/models/providers/anthropic/index.ts

@@ -0,0 +1,115 @@
+import { UIConfigField } from '@/lib/config/types';
+import { getConfiguredModelProviderById } from '@/lib/config/serverRegistry';
+import { Model, ModelList, ProviderMetadata } from '../../types';
+import BaseEmbedding from '../../base/embedding';
+import BaseModelProvider from '../../base/provider';
+import BaseLLM from '../../base/llm';
+import AnthropicLLM from './anthropicLLM';
+
+interface AnthropicConfig {
+  apiKey: string;
+}
+
+const providerConfigFields: UIConfigField[] = [
+  {
+    type: 'password',
+    name: 'API Key',
+    key: 'apiKey',
+    description: 'Your Anthropic API key',
+    required: true,
+    placeholder: 'Anthropic API Key',
+    env: 'ANTHROPIC_API_KEY',
+    scope: 'server',
+  },
+];
+
+class AnthropicProvider extends BaseModelProvider<AnthropicConfig> {
+  constructor(id: string, name: string, config: AnthropicConfig) {
+    super(id, name, config);
+  }
+
+  async getDefaultModels(): Promise<ModelList> {
+    const res = await fetch('https://api.anthropic.com/v1/models?limit=999', {
+      method: 'GET',
+      headers: {
+        'x-api-key': this.config.apiKey,
+        'anthropic-version': '2023-06-01',
+        'Content-type': 'application/json',
+      },
+    });
+
+    if (!res.ok) {
+      throw new Error(`Failed to fetch Anthropic models: ${res.statusText}`);
+    }
+
+    const data = (await res.json()).data;
+
+    const models: Model[] = data.map((m: any) => {
+      return {
+        key: m.id,
+        name: m.display_name,
+      };
+    });
+
+    return {
+      embedding: [],
+      chat: models,
+    };
+  }
+
+  async getModelList(): Promise<ModelList> {
+    const defaultModels = await this.getDefaultModels();
+    const configProvider = getConfiguredModelProviderById(this.id)!;
+
+    return {
+      embedding: [],
+      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 Anthropic Chat Model. Invalid Model Selected',
+      );
+    }
+
+    return new AnthropicLLM({
+      apiKey: this.config.apiKey,
+      model: key,
+      baseURL: 'https://api.anthropic.com/v1',
+    });
+  }
+
+  async loadEmbeddingModel(key: string): Promise<BaseEmbedding<any>> {
+    throw new Error('Anthropic provider does not support embedding models.');
+  }
+
+  static parseAndValidate(raw: any): AnthropicConfig {
+    if (!raw || typeof raw !== 'object')
+      throw new Error('Invalid config provided. Expected object');
+    if (!raw.apiKey)
+      throw new Error('Invalid config provided. API key must be provided');
+
+    return {
+      apiKey: String(raw.apiKey),
+    };
+  }
+
+  static getProviderConfigFields(): UIConfigField[] {
+    return providerConfigFields;
+  }
+
+  static getProviderMetadata(): ProviderMetadata {
+    return {
+      key: 'anthropic',
+      name: 'Anthropic',
+    };
+  }
+}
+
+export default AnthropicProvider;

apps/llm-proxy/src/lib/models/providers/gemini/geminiEmbedding.ts

@@ -0,0 +1,5 @@
+import OpenAIEmbedding from '../openai/openaiEmbedding';
+
+class GeminiEmbedding extends OpenAIEmbedding {}
+
+export default GeminiEmbedding;

apps/llm-proxy/src/lib/models/providers/gemini/geminiLLM.ts

@@ -0,0 +1,5 @@
+import OpenAILLM from '../openai/openaiLLM';
+
+class GeminiLLM extends OpenAILLM {}
+
+export default GeminiLLM;

apps/llm-proxy/src/lib/models/providers/gemini/index.ts

@@ -0,0 +1,144 @@
+import { UIConfigField } from '@/lib/config/types';
+import { getConfiguredModelProviderById } from '@/lib/config/serverRegistry';
+import { Model, ModelList, ProviderMetadata } from '../../types';
+import GeminiEmbedding from './geminiEmbedding';
+import BaseEmbedding from '../../base/embedding';
+import BaseModelProvider from '../../base/provider';
+import BaseLLM from '../../base/llm';
+import GeminiLLM from './geminiLLM';
+
+interface GeminiConfig {
+  apiKey: string;
+}
+
+const providerConfigFields: UIConfigField[] = [
+  {
+    type: 'password',
+    name: 'API Key',
+    key: 'apiKey',
+    description: 'Your Gemini API key',
+    required: true,
+    placeholder: 'Gemini API Key',
+    env: 'GEMINI_API_KEY',
+    scope: 'server',
+  },
+];
+
+class GeminiProvider extends BaseModelProvider<GeminiConfig> {
+  constructor(id: string, name: string, config: GeminiConfig) {
+    super(id, name, config);
+  }
+
+  async getDefaultModels(): Promise<ModelList> {
+    const res = await fetch(
+      `https://generativelanguage.googleapis.com/v1beta/models?key=${this.config.apiKey}`,
+      {
+        method: 'GET',
+        headers: {
+          'Content-Type': 'application/json',
+        },
+      },
+    );
+
+    const data = await res.json();
+
+    let defaultEmbeddingModels: Model[] = [];
+    let defaultChatModels: Model[] = [];
+
+    data.models.forEach((m: any) => {
+      if (
+        m.supportedGenerationMethods.some(
+          (genMethod: string) =>
+            genMethod === 'embedText' || genMethod === 'embedContent',
+        )
+      ) {
+        defaultEmbeddingModels.push({
+          key: m.name,
+          name: m.displayName,
+        });
+      } else if (m.supportedGenerationMethods.includes('generateContent')) {
+        defaultChatModels.push({
+          key: m.name,
+          name: m.displayName,
+        });
+      }
+    });
+
+    return {
+      embedding: defaultEmbeddingModels,
+      chat: defaultChatModels,
+    };
+  }
+
+  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 Gemini Chat Model. Invalid Model Selected',
+      );
+    }
+
+    return new GeminiLLM({
+      apiKey: this.config.apiKey,
+      model: key,
+      baseURL: 'https://generativelanguage.googleapis.com/v1beta/openai',
+    });
+  }
+
+  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 Gemini Embedding Model. Invalid Model Selected.',
+      );
+    }
+
+    return new GeminiEmbedding({
+      apiKey: this.config.apiKey,
+      model: key,
+      baseURL: 'https://generativelanguage.googleapis.com/v1beta/openai',
+    });
+  }
+
+  static parseAndValidate(raw: any): GeminiConfig {
+    if (!raw || typeof raw !== 'object')
+      throw new Error('Invalid config provided. Expected object');
+    if (!raw.apiKey)
+      throw new Error('Invalid config provided. API key must be provided');
+
+    return {
+      apiKey: String(raw.apiKey),
+    };
+  }
+
+  static getProviderConfigFields(): UIConfigField[] {
+    return providerConfigFields;
+  }
+
+  static getProviderMetadata(): ProviderMetadata {
+    return {
+      key: 'gemini',
+      name: 'Gemini',
+    };
+  }
+}
+
+export default GeminiProvider;

apps/llm-proxy/src/lib/models/providers/groq/groqLLM.ts

@@ -0,0 +1,5 @@
+import OpenAILLM from '../openai/openaiLLM';
+
+class GroqLLM extends OpenAILLM {}
+
+export default GroqLLM;

apps/llm-proxy/src/lib/models/providers/groq/index.ts

@@ -0,0 +1,113 @@
+import { UIConfigField } from '@/lib/config/types';
+import { getConfiguredModelProviderById } from '@/lib/config/serverRegistry';
+import { Model, ModelList, ProviderMetadata } from '../../types';
+import BaseEmbedding from '../../base/embedding';
+import BaseModelProvider from '../../base/provider';
+import BaseLLM from '../../base/llm';
+import GroqLLM from './groqLLM';
+
+interface GroqConfig {
+  apiKey: string;
+}
+
+const providerConfigFields: UIConfigField[] = [
+  {
+    type: 'password',
+    name: 'API Key',
+    key: 'apiKey',
+    description: 'Your Groq API key',
+    required: true,
+    placeholder: 'Groq API Key',
+    env: 'GROQ_API_KEY',
+    scope: 'server',
+  },
+];
+
+class GroqProvider extends BaseModelProvider<GroqConfig> {
+  constructor(id: string, name: string, config: GroqConfig) {
+    super(id, name, config);
+  }
+
+  async getDefaultModels(): Promise<ModelList> {
+    const res = await fetch(`https://api.groq.com/openai/v1/models`, {
+      method: 'GET',
+      headers: {
+        'Content-Type': 'application/json',
+        Authorization: `Bearer ${this.config.apiKey}`,
+      },
+    });
+
+    const data = await res.json();
+
+    const defaultChatModels: Model[] = [];
+
+    data.data.forEach((m: any) => {
+      defaultChatModels.push({
+        key: m.id,
+        name: m.id,
+      });
+    });
+
+    return {
+      embedding: [],
+      chat: defaultChatModels,
+    };
+  }
+
+  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 Groq Chat Model. Invalid Model Selected');
+    }
+
+    return new GroqLLM({
+      apiKey: this.config.apiKey,
+      model: key,
+      baseURL: 'https://api.groq.com/openai/v1',
+    });
+  }
+
+  async loadEmbeddingModel(key: string): Promise<BaseEmbedding<any>> {
+    throw new Error('Groq Provider does not support embedding models.');
+  }
+
+  static parseAndValidate(raw: any): GroqConfig {
+    if (!raw || typeof raw !== 'object')
+      throw new Error('Invalid config provided. Expected object');
+    if (!raw.apiKey)
+      throw new Error('Invalid config provided. API key must be provided');
+
+    return {
+      apiKey: String(raw.apiKey),
+    };
+  }
+
+  static getProviderConfigFields(): UIConfigField[] {
+    return providerConfigFields;
+  }
+
+  static getProviderMetadata(): ProviderMetadata {
+    return {
+      key: 'groq',
+      name: 'Groq',
+    };
+  }
+}
+
+export default GroqProvider;

apps/llm-proxy/src/lib/models/providers/index.ts

@@ -0,0 +1,35 @@
+import { ModelProviderUISection } from '@/lib/config/types';
+import { ProviderConstructor } from '../base/provider';
+import OpenAIProvider from './openai';
+import OllamaProvider from './ollama';
+import GeminiProvider from './gemini';
+import TransformersProvider from './transformers';
+import GroqProvider from './groq';
+import LemonadeProvider from './lemonade';
+import AnthropicProvider from './anthropic';
+import LMStudioProvider from './lmstudio';
+
+export const providers: Record<string, ProviderConstructor<any>> = {
+  openai: OpenAIProvider,
+  ollama: OllamaProvider,
+  gemini: GeminiProvider,
+  transformers: TransformersProvider,
+  groq: GroqProvider,
+  lemonade: LemonadeProvider,
+  anthropic: AnthropicProvider,
+  lmstudio: LMStudioProvider,
+};
+
+export const getModelProvidersUIConfigSection =
+  (): ModelProviderUISection[] => {
+    return Object.entries(providers).map(([k, p]) => {
+      const configFields = p.getProviderConfigFields();
+      const metadata = p.getProviderMetadata();
+
+      return {
+        fields: configFields,
+        key: k,
+        name: metadata.name,
+      };
+    });
+  };

apps/llm-proxy/src/lib/models/providers/lemonade/index.ts

@@ -0,0 +1,153 @@
+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 LemonadeLLM from './lemonadeLLM';
+import BaseEmbedding from '../../base/embedding';
+import LemonadeEmbedding from './lemonadeEmbedding';
+
+interface LemonadeConfig {
+  baseURL: string;
+  apiKey?: string;
+}
+
+const providerConfigFields: UIConfigField[] = [
+  {
+    type: 'string',
+    name: 'Base URL',
+    key: 'baseURL',
+    description: 'The base URL for Lemonade API',
+    required: true,
+    placeholder: 'https://api.lemonade.ai/v1',
+    env: 'LEMONADE_BASE_URL',
+    scope: 'server',
+  },
+  {
+    type: 'password',
+    name: 'API Key',
+    key: 'apiKey',
+    description: 'Your Lemonade API key (optional)',
+    required: false,
+    placeholder: 'Lemonade API Key',
+    env: 'LEMONADE_API_KEY',
+    scope: 'server',
+  },
+];
+
+class LemonadeProvider extends BaseModelProvider<LemonadeConfig> {
+  constructor(id: string, name: string, config: LemonadeConfig) {
+    super(id, name, config);
+  }
+
+  async getDefaultModels(): Promise<ModelList> {
+    try {
+      const res = await fetch(`${this.config.baseURL}/models`, {
+        method: 'GET',
+        headers: {
+          'Content-Type': 'application/json',
+          ...(this.config.apiKey
+            ? { Authorization: `Bearer ${this.config.apiKey}` }
+            : {}),
+        },
+      });
+
+      const data = await res.json();
+
+      const models: Model[] = data.data
+        .filter((m: any) => m.recipe === 'llamacpp')
+        .map((m: any) => {
+          return {
+            name: m.id,
+            key: m.id,
+          };
+        });
+
+      return {
+        embedding: models,
+        chat: models,
+      };
+    } catch (err) {
+      if (err instanceof TypeError) {
+        throw new Error(
+          'Error connecting to Lemonade API. Please ensure the base URL is correct and the service is available.',
+        );
+      }
+
+      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 Lemonade Chat Model. Invalid Model Selected',
+      );
+    }
+
+    return new LemonadeLLM({
+      apiKey: this.config.apiKey || 'not-needed',
+      model: key,
+      baseURL: this.config.baseURL,
+    });
+  }
+
+  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 Lemonade Embedding Model. Invalid Model Selected.',
+      );
+    }
+
+    return new LemonadeEmbedding({
+      apiKey: this.config.apiKey || 'not-needed',
+      model: key,
+      baseURL: this.config.baseURL,
+    });
+  }
+
+  static parseAndValidate(raw: any): LemonadeConfig {
+    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),
+      apiKey: raw.apiKey ? String(raw.apiKey) : undefined,
+    };
+  }
+
+  static getProviderConfigFields(): UIConfigField[] {
+    return providerConfigFields;
+  }
+
+  static getProviderMetadata(): ProviderMetadata {
+    return {
+      key: 'lemonade',
+      name: 'Lemonade',
+    };
+  }
+}
+
+export default LemonadeProvider;

apps/llm-proxy/src/lib/models/providers/lemonade/lemonadeEmbedding.ts

@@ -0,0 +1,5 @@
+import OpenAIEmbedding from '../openai/openaiEmbedding';
+
+class LemonadeEmbedding extends OpenAIEmbedding {}
+
+export default LemonadeEmbedding;

apps/llm-proxy/src/lib/models/providers/lemonade/lemonadeLLM.ts

@@ -0,0 +1,5 @@
+import OpenAILLM from '../openai/openaiLLM';
+
+class LemonadeLLM extends OpenAILLM {}
+
+export default LemonadeLLM;

apps/llm-proxy/src/lib/models/providers/lmstudio/index.ts

@@ -0,0 +1,143 @@
+import { UIConfigField } from '@/lib/config/types';
+import { getConfiguredModelProviderById } from '@/lib/config/serverRegistry';
+import BaseModelProvider from '../../base/provider';
+import { Model, ModelList, ProviderMetadata } from '../../types';
+import LMStudioLLM from './lmstudioLLM';
+import BaseLLM from '../../base/llm';
+import BaseEmbedding from '../../base/embedding';
+import LMStudioEmbedding from './lmstudioEmbedding';
+
+interface LMStudioConfig {
+  baseURL: string;
+}
+
+const providerConfigFields: UIConfigField[] = [
+  {
+    type: 'string',
+    name: 'Base URL',
+    key: 'baseURL',
+    description: 'The base URL for LM Studio server',
+    required: true,
+    placeholder: 'http://localhost:1234',
+    env: 'LM_STUDIO_BASE_URL',
+    scope: 'server',
+  },
+];
+
+class LMStudioProvider extends BaseModelProvider<LMStudioConfig> {
+  constructor(id: string, name: string, config: LMStudioConfig) {
+    super(id, name, config);
+  }
+
+  private normalizeBaseURL(url: string): string {
+    const trimmed = url.trim().replace(/\/+$/, '');
+    return trimmed.endsWith('/v1') ? trimmed : `${trimmed}/v1`;
+  }
+
+  async getDefaultModels(): Promise<ModelList> {
+    try {
+      const baseURL = this.normalizeBaseURL(this.config.baseURL);
+
+      const res = await fetch(`${baseURL}/models`, {
+        method: 'GET',
+        headers: {
+          'Content-Type': 'application/json',
+        },
+      });
+
+      const data = await res.json();
+
+      const models: Model[] = data.data.map((m: any) => {
+        return {
+          name: m.id,
+          key: m.id,
+        };
+      });
+
+      return {
+        embedding: models,
+        chat: models,
+      };
+    } catch (err) {
+      if (err instanceof TypeError) {
+        throw new Error(
+          'Error connecting to LM Studio. Please ensure the base URL is correct and the LM Studio 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 LM Studio Chat Model. Invalid Model Selected',
+      );
+    }
+
+    return new LMStudioLLM({
+      apiKey: 'lm-studio',
+      model: key,
+      baseURL: this.normalizeBaseURL(this.config.baseURL),
+    });
+  }
+
+  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 LM Studio Embedding Model. Invalid Model Selected.',
+      );
+    }
+
+    return new LMStudioEmbedding({
+      apiKey: 'lm-studio',
+      model: key,
+      baseURL: this.normalizeBaseURL(this.config.baseURL),
+    });
+  }
+
+  static parseAndValidate(raw: any): LMStudioConfig {
+    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: 'lmstudio',
+      name: 'LM Studio',
+    };
+  }
+}
+
+export default LMStudioProvider;

apps/llm-proxy/src/lib/models/providers/lmstudio/lmstudioEmbedding.ts

@@ -0,0 +1,5 @@
+import OpenAIEmbedding from '../openai/openaiEmbedding';
+
+class LMStudioEmbedding extends OpenAIEmbedding {}
+
+export default LMStudioEmbedding;

apps/llm-proxy/src/lib/models/providers/lmstudio/lmstudioLLM.ts

@@ -0,0 +1,5 @@
+import OpenAILLM from '../openai/openaiLLM';
+
+class LMStudioLLM extends OpenAILLM {}
+
+export default LMStudioLLM;

apps/llm-proxy/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;

apps/llm-proxy/src/lib/models/providers/ollama/ollamaEmbedding.ts

@@ -0,0 +1,40 @@
+import { Ollama } from 'ollama';
+import BaseEmbedding from '../../base/embedding';
+import { Chunk } from '@/lib/types';
+
+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;

apps/llm-proxy/src/lib/models/providers/ollama/ollamaLLM.ts

@@ -0,0 +1,261 @@
+import z from 'zod';
+import BaseLLM from '../../base/llm';
+import {
+  GenerateObjectInput,
+  GenerateOptions,
+  GenerateTextInput,
+  GenerateTextOutput,
+  StreamTextOutput,
+} from '../../types';
+import { Ollama, Tool as OllamaTool, Message as OllamaMessage } from 'ollama';
+import { parse } from 'partial-json';
+import crypto from 'crypto';
+import { Message } from '@/lib/types';
+import { repairJson } from '@toolsycc/json-repair';
+
+type OllamaConfig = {
+  baseURL: string;
+  model: string;
+  options?: GenerateOptions;
+};
+
+const reasoningModels = [
+  'gpt-oss',
+  'deepseek-r1',
+  'qwen3',
+  'deepseek-v3.1',
+  'magistral',
+  'nemotron-3-nano',
+];
+
+class OllamaLLM extends BaseLLM<OllamaConfig> {
+  ollamaClient: Ollama;
+
+  constructor(protected config: OllamaConfig) {
+    super(config);
+
+    this.ollamaClient = new Ollama({
+      host: this.config.baseURL || 'http://localhost:11434',
+    });
+  }
+
+  convertToOllamaMessages(messages: Message[]): OllamaMessage[] {
+    return messages.map((msg) => {
+      if (msg.role === 'tool') {
+        return {
+          role: 'tool',
+          tool_name: msg.name,
+          content: msg.content,
+        } as OllamaMessage;
+      } else if (msg.role === 'assistant') {
+        return {
+          role: 'assistant',
+          content: msg.content,
+          tool_calls:
+            msg.tool_calls?.map((tc, i) => ({
+              function: {
+                index: i,
+                name: tc.name,
+                arguments: tc.arguments,
+              },
+            })) || [],
+        };
+      }
+
+      return msg;
+    });
+  }
+
+  async generateText(input: GenerateTextInput): Promise<GenerateTextOutput> {
+    const ollamaTools: OllamaTool[] = [];
+
+    input.tools?.forEach((tool) => {
+      ollamaTools.push({
+        type: 'function',
+        function: {
+          name: tool.name,
+          description: tool.description,
+          parameters: z.toJSONSchema(tool.schema).properties,
+        },
+      });
+    });
+
+    const res = await this.ollamaClient.chat({
+      model: this.config.model,
+      messages: this.convertToOllamaMessages(input.messages),
+      tools: ollamaTools.length > 0 ? ollamaTools : undefined,
+      ...(reasoningModels.find((m) => this.config.model.includes(m))
+        ? { think: false }
+        : {}),
+      options: {
+        top_p: input.options?.topP ?? this.config.options?.topP,
+        temperature:
+          input.options?.temperature ?? this.config.options?.temperature ?? 0.7,
+        num_predict: input.options?.maxTokens ?? this.config.options?.maxTokens,
+        num_ctx: 32000,
+        frequency_penalty:
+          input.options?.frequencyPenalty ??
+          this.config.options?.frequencyPenalty,
+        presence_penalty:
+          input.options?.presencePenalty ??
+          this.config.options?.presencePenalty,
+        stop:
+          input.options?.stopSequences ?? this.config.options?.stopSequences,
+      },
+    });
+
+    return {
+      content: res.message.content,
+      toolCalls:
+        res.message.tool_calls?.map((tc) => ({
+          id: crypto.randomUUID(),
+          name: tc.function.name,
+          arguments: tc.function.arguments,
+        })) || [],
+      additionalInfo: {
+        reasoning: res.message.thinking,
+      },
+    };
+  }
+
+  async *streamText(
+    input: GenerateTextInput,
+  ): AsyncGenerator<StreamTextOutput> {
+    const ollamaTools: OllamaTool[] = [];
+
+    input.tools?.forEach((tool) => {
+      ollamaTools.push({
+        type: 'function',
+        function: {
+          name: tool.name,
+          description: tool.description,
+          parameters: z.toJSONSchema(tool.schema) as any,
+        },
+      });
+    });
+
+    const stream = await this.ollamaClient.chat({
+      model: this.config.model,
+      messages: this.convertToOllamaMessages(input.messages),
+      stream: true,
+      ...(reasoningModels.find((m) => this.config.model.includes(m))
+        ? { think: false }
+        : {}),
+      tools: ollamaTools.length > 0 ? ollamaTools : undefined,
+      options: {
+        top_p: input.options?.topP ?? this.config.options?.topP,
+        temperature:
+          input.options?.temperature ?? this.config.options?.temperature ?? 0.7,
+        num_ctx: 32000,
+        num_predict: input.options?.maxTokens ?? this.config.options?.maxTokens,
+        frequency_penalty:
+          input.options?.frequencyPenalty ??
+          this.config.options?.frequencyPenalty,
+        presence_penalty:
+          input.options?.presencePenalty ??
+          this.config.options?.presencePenalty,
+        stop:
+          input.options?.stopSequences ?? this.config.options?.stopSequences,
+      },
+    });
+
+    for await (const chunk of stream) {
+      yield {
+        contentChunk: chunk.message.content,
+        toolCallChunk:
+          chunk.message.tool_calls?.map((tc, i) => ({
+            id: crypto
+              .createHash('sha256')
+              .update(
+                `${i}-${tc.function.name}`,
+              ) /* Ollama currently doesn't return a tool call ID so we're creating one based on the index and tool call name */
+              .digest('hex'),
+            name: tc.function.name,
+            arguments: tc.function.arguments,
+          })) || [],
+        done: chunk.done,
+        additionalInfo: {
+          reasoning: chunk.message.thinking,
+        },
+      };
+    }
+  }
+
+  async generateObject<T>(input: GenerateObjectInput): Promise<T> {
+    const response = await this.ollamaClient.chat({
+      model: this.config.model,
+      messages: this.convertToOllamaMessages(input.messages),
+      format: z.toJSONSchema(input.schema),
+      ...(reasoningModels.find((m) => this.config.model.includes(m))
+        ? { think: false }
+        : {}),
+      options: {
+        top_p: input.options?.topP ?? this.config.options?.topP,
+        temperature:
+          input.options?.temperature ?? this.config.options?.temperature ?? 0.7,
+        num_predict: input.options?.maxTokens ?? this.config.options?.maxTokens,
+        frequency_penalty:
+          input.options?.frequencyPenalty ??
+          this.config.options?.frequencyPenalty,
+        presence_penalty:
+          input.options?.presencePenalty ??
+          this.config.options?.presencePenalty,
+        stop:
+          input.options?.stopSequences ?? this.config.options?.stopSequences,
+      },
+    });
+
+    try {
+      return input.schema.parse(
+        JSON.parse(
+          repairJson(response.message.content, {
+            extractJson: true,
+          }) as string,
+        ),
+      ) as T;
+    } catch (err) {
+      throw new Error(`Error parsing response from Ollama: ${err}`);
+    }
+  }
+
+  async *streamObject<T>(input: GenerateObjectInput): AsyncGenerator<T> {
+    let recievedObj: string = '';
+
+    const stream = await this.ollamaClient.chat({
+      model: this.config.model,
+      messages: this.convertToOllamaMessages(input.messages),
+      format: z.toJSONSchema(input.schema),
+      stream: true,
+      ...(reasoningModels.find((m) => this.config.model.includes(m))
+        ? { think: false }
+        : {}),
+      options: {
+        top_p: input.options?.topP ?? this.config.options?.topP,
+        temperature:
+          input.options?.temperature ?? this.config.options?.temperature ?? 0.7,
+        num_predict: input.options?.maxTokens ?? this.config.options?.maxTokens,
+        frequency_penalty:
+          input.options?.frequencyPenalty ??
+          this.config.options?.frequencyPenalty,
+        presence_penalty:
+          input.options?.presencePenalty ??
+          this.config.options?.presencePenalty,
+        stop:
+          input.options?.stopSequences ?? 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;

apps/llm-proxy/src/lib/models/providers/openai/index.ts

@@ -0,0 +1,226 @@
+import { UIConfigField } from '@/lib/config/types';
+import { getConfiguredModelProviderById } from '@/lib/config/serverRegistry';
+import { Model, ModelList, ProviderMetadata } from '../../types';
+import OpenAIEmbedding from './openaiEmbedding';
+import BaseEmbedding from '../../base/embedding';
+import BaseModelProvider from '../../base/provider';
+import BaseLLM from '../../base/llm';
+import OpenAILLM from './openaiLLM';
+
+interface OpenAIConfig {
+  apiKey: string;
+  baseURL: string;
+}
+
+const defaultChatModels: Model[] = [
+  {
+    name: 'GPT-3.5 Turbo',
+    key: 'gpt-3.5-turbo',
+  },
+  {
+    name: 'GPT-4',
+    key: 'gpt-4',
+  },
+  {
+    name: 'GPT-4 turbo',
+    key: 'gpt-4-turbo',
+  },
+  {
+    name: 'GPT-4 omni',
+    key: 'gpt-4o',
+  },
+  {
+    name: 'GPT-4o (2024-05-13)',
+    key: 'gpt-4o-2024-05-13',
+  },
+  {
+    name: 'GPT-4 omni mini',
+    key: 'gpt-4o-mini',
+  },
+  {
+    name: 'GPT 4.1 nano',
+    key: 'gpt-4.1-nano',
+  },
+  {
+    name: 'GPT 4.1 mini',
+    key: 'gpt-4.1-mini',
+  },
+  {
+    name: 'GPT 4.1',
+    key: 'gpt-4.1',
+  },
+  {
+    name: 'GPT 5 nano',
+    key: 'gpt-5-nano',
+  },
+  {
+    name: 'GPT 5',
+    key: 'gpt-5',
+  },
+  {
+    name: 'GPT 5 Mini',
+    key: 'gpt-5-mini',
+  },
+  {
+    name: 'GPT 5 Pro',
+    key: 'gpt-5-pro',
+  },
+  {
+    name: 'GPT 5.1',
+    key: 'gpt-5.1',
+  },
+  {
+    name: 'GPT 5.2',
+    key: 'gpt-5.2',
+  },
+  {
+    name: 'GPT 5.2 Pro',
+    key: 'gpt-5.2-pro',
+  },
+  {
+    name: 'o1',
+    key: 'o1',
+  },
+  {
+    name: 'o3',
+    key: 'o3',
+  },
+  {
+    name: 'o3 Mini',
+    key: 'o3-mini',
+  },
+  {
+    name: 'o4 Mini',
+    key: 'o4-mini',
+  },
+];
+
+const defaultEmbeddingModels: Model[] = [
+  {
+    name: 'Text Embedding 3 Small',
+    key: 'text-embedding-3-small',
+  },
+  {
+    name: 'Text Embedding 3 Large',
+    key: 'text-embedding-3-large',
+  },
+];
+
+const providerConfigFields: UIConfigField[] = [
+  {
+    type: 'password',
+    name: 'API Key',
+    key: 'apiKey',
+    description: 'Your OpenAI API key',
+    required: true,
+    placeholder: 'OpenAI API Key',
+    env: 'OPENAI_API_KEY',
+    scope: 'server',
+  },
+  {
+    type: 'string',
+    name: 'Base URL',
+    key: 'baseURL',
+    description: 'The base URL for the OpenAI API',
+    required: true,
+    placeholder: 'OpenAI Base URL',
+    default: 'https://api.openai.com/v1',
+    env: 'OPENAI_BASE_URL',
+    scope: 'server',
+  },
+];
+
+class OpenAIProvider extends BaseModelProvider<OpenAIConfig> {
+  constructor(id: string, name: string, config: OpenAIConfig) {
+    super(id, name, config);
+  }
+
+  async getDefaultModels(): Promise<ModelList> {
+    if (this.config.baseURL === 'https://api.openai.com/v1') {
+      return {
+        embedding: defaultEmbeddingModels,
+        chat: defaultChatModels,
+      };
+    }
+
+    return {
+      embedding: [],
+      chat: [],
+    };
+  }
+
+  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 OpenAI Chat Model. Invalid Model Selected',
+      );
+    }
+
+    return new OpenAILLM({
+      apiKey: this.config.apiKey,
+      model: key,
+      baseURL: this.config.baseURL,
+    });
+  }
+
+  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 OpenAI Embedding Model. Invalid Model Selected.',
+      );
+    }
+
+    return new OpenAIEmbedding({
+      apiKey: this.config.apiKey,
+      model: key,
+      baseURL: this.config.baseURL,
+    });
+  }
+
+  static parseAndValidate(raw: any): OpenAIConfig {
+    if (!raw || typeof raw !== 'object')
+      throw new Error('Invalid config provided. Expected object');
+    if (!raw.apiKey || !raw.baseURL)
+      throw new Error(
+        'Invalid config provided. API key and base URL must be provided',
+      );
+
+    return {
+      apiKey: String(raw.apiKey),
+      baseURL: String(raw.baseURL),
+    };
+  }
+
+  static getProviderConfigFields(): UIConfigField[] {
+    return providerConfigFields;
+  }
+
+  static getProviderMetadata(): ProviderMetadata {
+    return {
+      key: 'openai',
+      name: 'OpenAI',
+    };
+  }
+}
+
+export default OpenAIProvider;

apps/llm-proxy/src/lib/models/providers/openai/openaiEmbedding.ts

@@ -0,0 +1,42 @@
+import OpenAI from 'openai';
+import BaseEmbedding from '../../base/embedding';
+import { Chunk } from '@/lib/types';
+
+type OpenAIConfig = {
+  apiKey: string;
+  model: string;
+  baseURL?: string;
+};
+
+class OpenAIEmbedding extends BaseEmbedding<OpenAIConfig> {
+  openAIClient: OpenAI;
+
+  constructor(protected config: OpenAIConfig) {
+    super(config);
+
+    this.openAIClient = new OpenAI({
+      apiKey: config.apiKey,
+      baseURL: config.baseURL,
+    });
+  }
+
+  async embedText(texts: string[]): Promise<number[][]> {
+    const response = await this.openAIClient.embeddings.create({
+      model: this.config.model,
+      input: texts,
+    });
+
+    return response.data.map((embedding) => embedding.embedding);
+  }
+
+  async embedChunks(chunks: Chunk[]): Promise<number[][]> {
+    const response = await this.openAIClient.embeddings.create({
+      model: this.config.model,
+      input: chunks.map((c) => c.content),
+    });
+
+    return response.data.map((embedding) => embedding.embedding);
+  }
+}
+
+export default OpenAIEmbedding;

apps/llm-proxy/src/lib/models/providers/openai/openaiLLM.ts

@@ -0,0 +1,275 @@
+import OpenAI from 'openai';
+import BaseLLM from '../../base/llm';
+import { zodTextFormat, zodResponseFormat } from 'openai/helpers/zod';
+import {
+  GenerateObjectInput,
+  GenerateOptions,
+  GenerateTextInput,
+  GenerateTextOutput,
+  StreamTextOutput,
+  ToolCall,
+} from '../../types';
+import { parse } from 'partial-json';
+import z from 'zod';
+import {
+  ChatCompletionAssistantMessageParam,
+  ChatCompletionMessageParam,
+  ChatCompletionTool,
+  ChatCompletionToolMessageParam,
+} from 'openai/resources/index.mjs';
+import { Message } from '@/lib/types';
+import { repairJson } from '@toolsycc/json-repair';
+
+type OpenAIConfig = {
+  apiKey: string;
+  model: string;
+  baseURL?: string;
+  options?: GenerateOptions;
+};
+
+class OpenAILLM extends BaseLLM<OpenAIConfig> {
+  openAIClient: OpenAI;
+
+  constructor(protected config: OpenAIConfig) {
+    super(config);
+
+    this.openAIClient = new OpenAI({
+      apiKey: this.config.apiKey,
+      baseURL: this.config.baseURL || 'https://api.openai.com/v1',
+    });
+  }
+
+  convertToOpenAIMessages(messages: Message[]): ChatCompletionMessageParam[] {
+    return messages.map((msg) => {
+      if (msg.role === 'tool') {
+        return {
+          role: 'tool',
+          tool_call_id: msg.id,
+          content: msg.content,
+        } as ChatCompletionToolMessageParam;
+      } else if (msg.role === 'assistant') {
+        return {
+          role: 'assistant',
+          content: msg.content,
+          ...(msg.tool_calls &&
+            msg.tool_calls.length > 0 && {
+              tool_calls: msg.tool_calls?.map((tc) => ({
+                id: tc.id,
+                type: 'function',
+                function: {
+                  name: tc.name,
+                  arguments: JSON.stringify(tc.arguments),
+                },
+              })),
+            }),
+        } as ChatCompletionAssistantMessageParam;
+      }
+
+      return msg;
+    });
+  }
+
+  async generateText(input: GenerateTextInput): Promise<GenerateTextOutput> {
+    const openaiTools: ChatCompletionTool[] = [];
+
+    input.tools?.forEach((tool) => {
+      openaiTools.push({
+        type: 'function',
+        function: {
+          name: tool.name,
+          description: tool.description,
+          parameters: z.toJSONSchema(tool.schema),
+        },
+      });
+    });
+
+    const response = await this.openAIClient.chat.completions.create({
+      model: this.config.model,
+      tools: openaiTools.length > 0 ? openaiTools : undefined,
+      messages: this.convertToOpenAIMessages(input.messages),
+      temperature:
+        input.options?.temperature ?? this.config.options?.temperature ?? 1.0,
+      top_p: input.options?.topP ?? this.config.options?.topP,
+      max_completion_tokens:
+        input.options?.maxTokens ?? this.config.options?.maxTokens,
+      stop: input.options?.stopSequences ?? this.config.options?.stopSequences,
+      frequency_penalty:
+        input.options?.frequencyPenalty ??
+        this.config.options?.frequencyPenalty,
+      presence_penalty:
+        input.options?.presencePenalty ?? this.config.options?.presencePenalty,
+    });
+
+    if (response.choices && response.choices.length > 0) {
+      return {
+        content: response.choices[0].message.content!,
+        toolCalls:
+          response.choices[0].message.tool_calls
+            ?.map((tc) => {
+              if (tc.type === 'function') {
+                return {
+                  name: tc.function.name,
+                  id: tc.id,
+                  arguments: JSON.parse(tc.function.arguments),
+                };
+              }
+            })
+            .filter((tc) => tc !== undefined) || [],
+        additionalInfo: {
+          finishReason: response.choices[0].finish_reason,
+        },
+      };
+    }
+
+    throw new Error('No response from OpenAI');
+  }
+
+  async *streamText(
+    input: GenerateTextInput,
+  ): AsyncGenerator<StreamTextOutput> {
+    const openaiTools: ChatCompletionTool[] = [];
+
+    input.tools?.forEach((tool) => {
+      openaiTools.push({
+        type: 'function',
+        function: {
+          name: tool.name,
+          description: tool.description,
+          parameters: z.toJSONSchema(tool.schema),
+        },
+      });
+    });
+
+    const stream = await this.openAIClient.chat.completions.create({
+      model: this.config.model,
+      messages: this.convertToOpenAIMessages(input.messages),
+      tools: openaiTools.length > 0 ? openaiTools : undefined,
+      temperature:
+        input.options?.temperature ?? this.config.options?.temperature ?? 1.0,
+      top_p: input.options?.topP ?? this.config.options?.topP,
+      max_completion_tokens:
+        input.options?.maxTokens ?? this.config.options?.maxTokens,
+      stop: input.options?.stopSequences ?? this.config.options?.stopSequences,
+      frequency_penalty:
+        input.options?.frequencyPenalty ??
+        this.config.options?.frequencyPenalty,
+      presence_penalty:
+        input.options?.presencePenalty ?? this.config.options?.presencePenalty,
+      stream: true,
+    });
+
+    let recievedToolCalls: { name: string; id: string; arguments: string }[] =
+      [];
+
+    for await (const chunk of stream) {
+      if (chunk.choices && chunk.choices.length > 0) {
+        const toolCalls = chunk.choices[0].delta.tool_calls;
+        yield {
+          contentChunk: chunk.choices[0].delta.content || '',
+          toolCallChunk:
+            toolCalls?.map((tc) => {
+              if (!recievedToolCalls[tc.index]) {
+                const call = {
+                  name: tc.function?.name!,
+                  id: tc.id!,
+                  arguments: tc.function?.arguments || '',
+                };
+                recievedToolCalls.push(call);
+                return { ...call, arguments: parse(call.arguments || '{}') };
+              } else {
+                const existingCall = recievedToolCalls[tc.index];
+                existingCall.arguments += tc.function?.arguments || '';
+                return {
+                  ...existingCall,
+                  arguments: parse(existingCall.arguments),
+                };
+              }
+            }) || [],
+          done: chunk.choices[0].finish_reason !== null,
+          additionalInfo: {
+            finishReason: chunk.choices[0].finish_reason,
+          },
+        };
+      }
+    }
+  }
+
+  async generateObject<T>(input: GenerateObjectInput): Promise<T> {
+    const response = await this.openAIClient.chat.completions.parse({
+      messages: this.convertToOpenAIMessages(input.messages),
+      model: this.config.model,
+      temperature:
+        input.options?.temperature ?? this.config.options?.temperature ?? 1.0,
+      top_p: input.options?.topP ?? this.config.options?.topP,
+      max_completion_tokens:
+        input.options?.maxTokens ?? this.config.options?.maxTokens,
+      stop: input.options?.stopSequences ?? this.config.options?.stopSequences,
+      frequency_penalty:
+        input.options?.frequencyPenalty ??
+        this.config.options?.frequencyPenalty,
+      presence_penalty:
+        input.options?.presencePenalty ?? this.config.options?.presencePenalty,
+      response_format: zodResponseFormat(input.schema, 'object'),
+    });
+
+    if (response.choices && response.choices.length > 0) {
+      try {
+        return input.schema.parse(
+          JSON.parse(
+            repairJson(response.choices[0].message.content!, {
+              extractJson: true,
+            }) as string,
+          ),
+        ) as T;
+      } catch (err) {
+        throw new Error(`Error parsing response from OpenAI: ${err}`);
+      }
+    }
+
+    throw new Error('No response from OpenAI');
+  }
+
+  async *streamObject<T>(input: GenerateObjectInput): AsyncGenerator<T> {
+    let recievedObj: string = '';
+
+    const stream = this.openAIClient.responses.stream({
+      model: this.config.model,
+      input: input.messages,
+      temperature:
+        input.options?.temperature ?? this.config.options?.temperature ?? 1.0,
+      top_p: input.options?.topP ?? this.config.options?.topP,
+      max_completion_tokens:
+        input.options?.maxTokens ?? this.config.options?.maxTokens,
+      stop: input.options?.stopSequences ?? this.config.options?.stopSequences,
+      frequency_penalty:
+        input.options?.frequencyPenalty ??
+        this.config.options?.frequencyPenalty,
+      presence_penalty:
+        input.options?.presencePenalty ?? this.config.options?.presencePenalty,
+      text: {
+        format: zodTextFormat(input.schema, 'object'),
+      },
+    });
+
+    for await (const chunk of stream) {
+      if (chunk.type === 'response.output_text.delta' && chunk.delta) {
+        recievedObj += chunk.delta;
+
+        try {
+          yield parse(recievedObj) as T;
+        } catch (err) {
+          console.log('Error parsing partial object from OpenAI:', err);
+          yield {} as T;
+        }
+      } else if (chunk.type === 'response.output_text.done' && chunk.text) {
+        try {
+          yield parse(chunk.text) as T;
+        } catch (err) {
+          throw new Error(`Error parsing response from OpenAI: ${err}`);
+        }
+      }
+    }
+  }
+}
+
+export default OpenAILLM;

apps/llm-proxy/src/lib/models/providers/transformers/index.ts

@@ -0,0 +1,88 @@
+import { UIConfigField } from '@/lib/config/types';
+import { getConfiguredModelProviderById } from '@/lib/config/serverRegistry';
+import { Model, ModelList, ProviderMetadata } from '../../types';
+import BaseModelProvider from '../../base/provider';
+import BaseLLM from '../../base/llm';
+import BaseEmbedding from '../../base/embedding';
+import TransformerEmbedding from './transformerEmbedding';
+
+interface TransformersConfig {}
+
+const defaultEmbeddingModels: Model[] = [
+  {
+    name: 'all-MiniLM-L6-v2',
+    key: 'Xenova/all-MiniLM-L6-v2',
+  },
+  {
+    name: 'mxbai-embed-large-v1',
+    key: 'mixedbread-ai/mxbai-embed-large-v1',
+  },
+  {
+    name: 'nomic-embed-text-v1',
+    key: 'Xenova/nomic-embed-text-v1',
+  },
+];
+
+const providerConfigFields: UIConfigField[] = [];
+
+class TransformersProvider extends BaseModelProvider<TransformersConfig> {
+  constructor(id: string, name: string, config: TransformersConfig) {
+    super(id, name, config);
+  }
+
+  async getDefaultModels(): Promise<ModelList> {
+    return {
+      embedding: [...defaultEmbeddingModels],
+      chat: [],
+    };
+  }
+
+  async getModelList(): Promise<ModelList> {
+    const defaultModels = await this.getDefaultModels();
+    const configProvider = getConfiguredModelProviderById(this.id)!;
+
+    return {
+      embedding: [
+        ...defaultModels.embedding,
+        ...configProvider.embeddingModels,
+      ],
+      chat: [],
+    };
+  }
+
+  async loadChatModel(key: string): Promise<BaseLLM<any>> {
+    throw new Error('Transformers Provider does not support chat models.');
+  }
+
+  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 OpenAI Embedding Model. Invalid Model Selected.',
+      );
+    }
+
+    return new TransformerEmbedding({
+      model: key,
+    });
+  }
+
+  static parseAndValidate(raw: any): TransformersConfig {
+    return {};
+  }
+
+  static getProviderConfigFields(): UIConfigField[] {
+    return providerConfigFields;
+  }
+
+  static getProviderMetadata(): ProviderMetadata {
+    return {
+      key: 'transformers',
+      name: 'Transformers',
+    };
+  }
+}
+
+export default TransformersProvider;

apps/llm-proxy/src/lib/models/providers/transformers/transformerEmbedding.ts

@@ -0,0 +1,41 @@
+import { Chunk } from '@/lib/types';
+import BaseEmbedding from '../../base/embedding';
+import { FeatureExtractionPipeline } from '@huggingface/transformers';
+
+type TransformerConfig = {
+  model: string;
+};
+
+class TransformerEmbedding extends BaseEmbedding<TransformerConfig> {
+  private pipelinePromise: Promise<FeatureExtractionPipeline> | null = null;
+
+  constructor(protected config: TransformerConfig) {
+    super(config);
+  }
+
+  async embedText(texts: string[]): Promise<number[][]> {
+    return this.embed(texts);
+  }
+
+  async embedChunks(chunks: Chunk[]): Promise<number[][]> {
+    return this.embed(chunks.map((c) => c.content));
+  }
+
+  private async embed(texts: string[]) {
+    if (!this.pipelinePromise) {
+      this.pipelinePromise = (async () => {
+        const { pipeline } = await import('@huggingface/transformers');
+        const result = await pipeline('feature-extraction', this.config.model, {
+          dtype: 'fp32',
+        });
+        return result as FeatureExtractionPipeline;
+      })();
+    }
+
+    const pipe = await this.pipelinePromise;
+    const output = await pipe(texts, { pooling: 'mean', normalize: true });
+    return output.tolist() as number[][];
+  }
+}
+
+export default TransformerEmbedding;

apps/llm-proxy/src/lib/models/registry.ts

@@ -0,0 +1,221 @@
+import { ConfigModelProvider } from '../config/types';
+import BaseModelProvider, { createProviderInstance } from './base/provider';
+import { getConfiguredModelProviders } from '../config/serverRegistry';
+import { providers } from './providers';
+import { MinimalProvider, ModelList } from './types';
+import configManager from '../config';
+
+class ModelRegistry {
+  activeProviders: (ConfigModelProvider & {
+    provider: BaseModelProvider<any>;
+  })[] = [];
+
+  constructor() {
+    this.initializeActiveProviders();
+  }
+
+  private initializeActiveProviders() {
+    const configuredProviders = getConfiguredModelProviders();
+
+    configuredProviders.forEach((p) => {
+      try {
+        const provider = providers[p.type];
+        if (!provider) throw new Error('Invalid provider type');
+
+        this.activeProviders.push({
+          ...p,
+          provider: createProviderInstance(provider, p.id, p.name, p.config),
+        });
+      } catch (err) {
+        console.error(
+          `Failed to initialize provider. Type: ${p.type}, ID: ${p.id}, Config: ${JSON.stringify(p.config)}, Error: ${err}`,
+        );
+      }
+    });
+  }
+
+  async getActiveProviders() {
+    const providers: MinimalProvider[] = [];
+
+    await Promise.all(
+      this.activeProviders.map(async (p) => {
+        let m: ModelList = { chat: [], embedding: [] };
+
+        try {
+          m = await p.provider.getModelList();
+        } catch (err: any) {
+          console.error(
+            `Failed to get model list. Type: ${p.type}, ID: ${p.id}, Error: ${err.message}`,
+          );
+
+          m = {
+            chat: [
+              {
+                key: 'error',
+                name: err.message,
+              },
+            ],
+            embedding: [],
+          };
+        }
+
+        providers.push({
+          id: p.id,
+          name: p.name,
+          chatModels: m.chat,
+          embeddingModels: m.embedding,
+        });
+      }),
+    );
+
+    return providers;
+  }
+
+  async loadChatModel(providerId: string, modelName: string) {
+    const provider = this.activeProviders.find((p) => p.id === providerId);
+
+    if (!provider) throw new Error('Invalid provider id');
+
+    const model = await provider.provider.loadChatModel(modelName);
+
+    return model;
+  }
+
+  async loadEmbeddingModel(providerId: string, modelName: string) {
+    const provider = this.activeProviders.find((p) => p.id === providerId);
+
+    if (!provider) throw new Error('Invalid provider id');
+
+    const model = await provider.provider.loadEmbeddingModel(modelName);
+
+    return model;
+  }
+
+  async addProvider(
+    type: string,
+    name: string,
+    config: Record<string, any>,
+  ): Promise<ConfigModelProvider> {
+    const provider = providers[type];
+    if (!provider) throw new Error('Invalid provider type');
+
+    const newProvider = configManager.addModelProvider(type, name, config);
+
+    const instance = createProviderInstance(
+      provider,
+      newProvider.id,
+      newProvider.name,
+      newProvider.config,
+    );
+
+    let m: ModelList = { chat: [], embedding: [] };
+
+    try {
+      m = await instance.getModelList();
+    } catch (err: any) {
+      console.error(
+        `Failed to get model list for newly added provider. Type: ${type}, ID: ${newProvider.id}, Error: ${err.message}`,
+      );
+
+      m = {
+        chat: [
+          {
+            key: 'error',
+            name: err.message,
+          },
+        ],
+        embedding: [],
+      };
+    }
+
+    this.activeProviders.push({
+      ...newProvider,
+      provider: instance,
+    });
+
+    return {
+      ...newProvider,
+      chatModels: m.chat || [],
+      embeddingModels: m.embedding || [],
+    };
+  }
+
+  async removeProvider(providerId: string): Promise<void> {
+    configManager.removeModelProvider(providerId);
+    this.activeProviders = this.activeProviders.filter(
+      (p) => p.id !== providerId,
+    );
+
+    return;
+  }
+
+  async updateProvider(
+    providerId: string,
+    name: string,
+    config: any,
+  ): Promise<ConfigModelProvider> {
+    const updated = await configManager.updateModelProvider(
+      providerId,
+      name,
+      config,
+    );
+    const instance = createProviderInstance(
+      providers[updated.type],
+      providerId,
+      name,
+      config,
+    );
+
+    let m: ModelList = { chat: [], embedding: [] };
+
+    try {
+      m = await instance.getModelList();
+    } catch (err: any) {
+      console.error(
+        `Failed to get model list for updated provider. Type: ${updated.type}, ID: ${updated.id}, Error: ${err.message}`,
+      );
+
+      m = {
+        chat: [
+          {
+            key: 'error',
+            name: err.message,
+          },
+        ],
+        embedding: [],
+      };
+    }
+
+    this.activeProviders.push({
+      ...updated,
+      provider: instance,
+    });
+
+    return {
+      ...updated,
+      chatModels: m.chat || [],
+      embeddingModels: m.embedding || [],
+    };
+  }
+
+  /* Using async here because maybe in the future we might want to add some validation?? */
+  async addProviderModel(
+    providerId: string,
+    type: 'embedding' | 'chat',
+    model: any,
+  ): Promise<any> {
+    const addedModel = configManager.addProviderModel(providerId, type, model);
+    return addedModel;
+  }
+
+  async removeProviderModel(
+    providerId: string,
+    type: 'embedding' | 'chat',
+    modelKey: string,
+  ): Promise<void> {
+    configManager.removeProviderModel(providerId, type, modelKey);
+    return;
+  }
+}
+
+export default ModelRegistry;

apps/llm-proxy/src/lib/models/types.ts

@@ -0,0 +1,103 @@
+import z from 'zod';
+import { Message } from '../types';
+
+type Model = {
+  name: string;
+  key: string;
+};
+
+type ModelList = {
+  embedding: Model[];
+  chat: Model[];
+};
+
+type ProviderMetadata = {
+  name: string;
+  key: string;
+};
+
+type MinimalProvider = {
+  id: string;
+  name: string;
+  chatModels: Model[];
+  embeddingModels: Model[];
+};
+
+type ModelWithProvider = {
+  key: string;
+  providerId: string;
+};
+
+type GenerateOptions = {
+  temperature?: number;
+  maxTokens?: number;
+  topP?: number;
+  stopSequences?: string[];
+  frequencyPenalty?: number;
+  presencePenalty?: number;
+};
+
+type Tool = {
+  name: string;
+  description: string;
+  schema: z.ZodObject<any>;
+};
+
+type ToolCall = {
+  id: string;
+  name: string;
+  arguments: Record<string, any>;
+};
+
+type GenerateTextInput = {
+  messages: Message[];
+  tools?: Tool[];
+  options?: GenerateOptions;
+};
+
+type GenerateTextOutput = {
+  content: string;
+  toolCalls: ToolCall[];
+  additionalInfo?: Record<string, any>;
+};
+
+type StreamTextOutput = {
+  contentChunk: string;
+  toolCallChunk: ToolCall[];
+  additionalInfo?: Record<string, any>;
+  done?: boolean;
+};
+
+type GenerateObjectInput = {
+  schema: z.ZodTypeAny;
+  messages: Message[];
+  options?: GenerateOptions;
+};
+
+type GenerateObjectOutput<T> = {
+  object: T;
+  additionalInfo?: Record<string, any>;
+};
+
+type StreamObjectOutput<T> = {
+  objectChunk: Partial<T>;
+  additionalInfo?: Record<string, any>;
+  done?: boolean;
+};
+
+export type {
+  Model,
+  ModelList,
+  ProviderMetadata,
+  MinimalProvider,
+  ModelWithProvider,
+  GenerateOptions,
+  GenerateTextInput,
+  GenerateTextOutput,
+  StreamTextOutput,
+  GenerateObjectInput,
+  GenerateObjectOutput,
+  StreamObjectOutput,
+  Tool,
+  ToolCall,
+};