Compare commits
31 Commits
| Author | SHA1 | Date | |
|---|---|---|---|
| 0a80ac0a8a | |||
| 6ce442354e | |||
| 9b38a3c06e | |||
| 5dead05324 | |||
| 6916dd9e2a | |||
| f89888a542 | |||
| d93b198b09 | |||
| 9e390d0fdb | |||
| 8329ee861e | |||
| b8585a0afb | |||
| c96f5118cf | |||
| 17e1a1f1e1 | |||
| de940dff75 | |||
| 4fc1e029e4 | |||
| d0a4151a2b | |||
| ad5dd4799b | |||
| 1c49af74ac | |||
| eda8ce36df | |||
| e82c510094 | |||
| 0378308721 | |||
| 189a32683f | |||
| f731b9f78d | |||
| 3701e21284 | |||
| 490d4996d2 | |||
| f099a8f1ed | |||
| a0228a0abc | |||
| a5257b52e7 | |||
| a4144fc071 | |||
| af46b3e81e | |||
| d50427937c | |||
| ffde2e0bf1 |
130
changelog.md
Normal file
130
changelog.md
Normal file
@ -0,0 +1,130 @@
|
||||
# Changelog
|
||||
|
||||
## 2025-02-25 - 0.4.2 - fix(core)
|
||||
Fix OpenAI chat streaming and PDF document processing logic.
|
||||
|
||||
- Updated OpenAI chat streaming to handle new async iterable format.
|
||||
- Improved PDF document processing by filtering out empty image buffers.
|
||||
- Removed unsupported temperature options from OpenAI requests.
|
||||
|
||||
## 2025-02-25 - 0.4.1 - fix(provider)
|
||||
Fix provider modules for consistency
|
||||
|
||||
- Updated TypeScript interfaces and options in provider modules for better type safety.
|
||||
- Modified transform stream handlers in Exo, Groq, and Ollama providers for consistency.
|
||||
- Added optional model options to OpenAI provider for custom model usage.
|
||||
|
||||
## 2025-02-08 - 0.4.0 - feat(core)
|
||||
Added support for Exo AI provider
|
||||
|
||||
- Introduced ExoProvider with chat functionalities.
|
||||
- Updated SmartAi class to initialize ExoProvider.
|
||||
- Extended Conversation class to support ExoProvider.
|
||||
|
||||
## 2025-02-05 - 0.3.3 - fix(documentation)
|
||||
Update readme with detailed license and legal information.
|
||||
|
||||
- Added explicit section on License and Legal Information in the README.
|
||||
- Clarified the use of trademarks and company information.
|
||||
|
||||
## 2025-02-05 - 0.3.2 - fix(documentation)
|
||||
Remove redundant badges from readme
|
||||
|
||||
- Removed Build Status badge from the readme file.
|
||||
- Removed License badge from the readme file.
|
||||
|
||||
## 2025-02-05 - 0.3.1 - fix(documentation)
|
||||
Updated README structure and added detailed usage examples
|
||||
|
||||
- Introduced a Table of Contents
|
||||
- Included comprehensive sections for chat, streaming chat, audio generation, document processing, and vision processing
|
||||
- Added example code and detailed configuration steps for supported AI providers
|
||||
- Clarified the development setup with instructions for running tests and building the project
|
||||
|
||||
## 2025-02-05 - 0.3.0 - feat(integration-xai)
|
||||
Add support for X.AI provider with chat and document processing capabilities.
|
||||
|
||||
- Introduced XAIProvider class for integrating X.AI features.
|
||||
- Implemented chat streaming and synchronous chat for X.AI.
|
||||
- Enabled document processing capabilities with PDF conversion in X.AI.
|
||||
|
||||
## 2025-02-03 - 0.2.0 - feat(provider.anthropic)
|
||||
Add support for vision and document processing in Anthropic provider
|
||||
|
||||
- Implemented vision tasks for Anthropic provider using Claude-3-opus-20240229 model.
|
||||
- Implemented document processing for Anthropic provider, supporting conversion of PDF documents to images and analysis with Claude-3-opus-20240229 model.
|
||||
- Updated documentation to reflect the new capabilities of the Anthropic provider.
|
||||
|
||||
## 2025-02-03 - 0.1.0 - feat(providers)
|
||||
Add vision and document processing capabilities to providers
|
||||
|
||||
- OpenAI and Ollama providers now support vision tasks using GPT-4 Vision and Llava models respectively.
|
||||
- Document processing has been implemented for OpenAI and Ollama providers, converting PDFs to images for analysis.
|
||||
- Introduced abstract methods for vision and document processing in the MultiModalModel class.
|
||||
- Updated the readme file with examples for vision and document processing.
|
||||
|
||||
## 2025-02-03 - 0.0.19 - fix(core)
|
||||
Enhanced chat streaming and error handling across providers
|
||||
|
||||
- Refactored chatStream method to properly handle input streams and processes in Perplexity, OpenAI, Ollama, and Anthropic providers.
|
||||
- Improved error handling and message parsing in chatStream implementations.
|
||||
- Defined distinct interfaces for chat options, messages, and responses.
|
||||
- Adjusted the test logic in test/test.ts for the new classification response requirement.
|
||||
|
||||
## 2024-09-19 - 0.0.18 - fix(dependencies)
|
||||
Update dependencies to the latest versions.
|
||||
|
||||
- Updated @git.zone/tsbuild from ^2.1.76 to ^2.1.84
|
||||
- Updated @git.zone/tsrun from ^1.2.46 to ^1.2.49
|
||||
- Updated @push.rocks/tapbundle from ^5.0.23 to ^5.3.0
|
||||
- Updated @types/node from ^20.12.12 to ^22.5.5
|
||||
- Updated @anthropic-ai/sdk from ^0.21.0 to ^0.27.3
|
||||
- Updated @push.rocks/smartfile from ^11.0.14 to ^11.0.21
|
||||
- Updated @push.rocks/smartpromise from ^4.0.3 to ^4.0.4
|
||||
- Updated @push.rocks/webstream from ^1.0.8 to ^1.0.10
|
||||
- Updated openai from ^4.47.1 to ^4.62.1
|
||||
|
||||
## 2024-05-29 - 0.0.17 - Documentation
|
||||
Updated project description.
|
||||
|
||||
- Improved project description for clarity and details.
|
||||
|
||||
## 2024-05-17 - 0.0.16 to 0.0.15 - Core
|
||||
Fixes and updates.
|
||||
|
||||
- Various core updates and fixes for stability improvements.
|
||||
|
||||
## 2024-04-29 - 0.0.14 to 0.0.13 - Core
|
||||
Fixes and updates.
|
||||
|
||||
- Multiple core updates and fixes for enhanced functionality.
|
||||
|
||||
## 2024-04-29 - 0.0.12 - Core
|
||||
Fixes and updates.
|
||||
|
||||
- Core update and bug fixes.
|
||||
|
||||
## 2024-04-29 - 0.0.11 - Provider
|
||||
Fix integration for anthropic provider.
|
||||
|
||||
- Correction in the integration process with anthropic provider for better compatibility.
|
||||
|
||||
## 2024-04-27 - 0.0.10 to 0.0.9 - Core
|
||||
Fixes and updates.
|
||||
|
||||
- Updates and fixes to core components.
|
||||
- Updated tsconfig for improved TypeScript configuration.
|
||||
|
||||
## 2024-04-01 - 0.0.8 to 0.0.7 - Core and npmextra
|
||||
Core updates and npmextra configuration.
|
||||
|
||||
- Core fixes and updates.
|
||||
- Updates to npmextra.json for githost configuration.
|
||||
|
||||
## 2024-03-31 - 0.0.6 to 0.0.2 - Core
|
||||
Initial core updates and fixes.
|
||||
|
||||
- Multiple updates and fixes to core following initial versions.
|
||||
|
||||
|
||||
This summarizes the relevant updates and changes based on the provided commit messages. The changelog excludes commits that are version tags without meaningful content or repeated entries.
|
||||
19
license
Normal file
19
license
Normal file
@ -0,0 +1,19 @@
|
||||
Copyright (c) 2024 Task Venture Capital GmbH (hello@task.vc)
|
||||
|
||||
Permission is hereby granted, free of charge, to any person obtaining a copy
|
||||
of this software and associated documentation files (the "Software"), to deal
|
||||
in the Software without restriction, including without limitation the rights
|
||||
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
||||
copies of the Software, and to permit persons to whom the Software is
|
||||
furnished to do so, subject to the following conditions:
|
||||
|
||||
The above copyright notice and this permission notice shall be included in all
|
||||
copies or substantial portions of the Software.
|
||||
|
||||
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
||||
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
||||
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
||||
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
||||
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
||||
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
||||
SOFTWARE.
|
||||
41
package.json
41
package.json
@ -1,6 +1,6 @@
|
||||
{
|
||||
"name": "@push.rocks/smartai",
|
||||
"version": "0.0.13",
|
||||
"version": "0.4.2",
|
||||
"private": false,
|
||||
"description": "A TypeScript library for integrating and interacting with multiple AI models, offering capabilities for chat and potentially audio responses.",
|
||||
"main": "dist_ts/index.js",
|
||||
@ -14,33 +14,33 @@
|
||||
"buildDocs": "(tsdoc)"
|
||||
},
|
||||
"devDependencies": {
|
||||
"@git.zone/tsbuild": "^2.1.25",
|
||||
"@git.zone/tsbundle": "^2.0.5",
|
||||
"@git.zone/tsrun": "^1.2.46",
|
||||
"@git.zone/tstest": "^1.0.90",
|
||||
"@push.rocks/qenv": "^6.0.5",
|
||||
"@push.rocks/tapbundle": "^5.0.23",
|
||||
"@types/node": "^20.12.7"
|
||||
"@git.zone/tsbuild": "^2.2.1",
|
||||
"@git.zone/tsbundle": "^2.2.5",
|
||||
"@git.zone/tsrun": "^1.3.3",
|
||||
"@git.zone/tstest": "^1.0.96",
|
||||
"@push.rocks/qenv": "^6.1.0",
|
||||
"@push.rocks/tapbundle": "^5.5.6",
|
||||
"@types/node": "^22.13.5"
|
||||
},
|
||||
"dependencies": {
|
||||
"@anthropic-ai/sdk": "^0.20.7",
|
||||
"@push.rocks/smartarray": "^1.0.8",
|
||||
"@push.rocks/smartfile": "^11.0.14",
|
||||
"@anthropic-ai/sdk": "^0.37.0",
|
||||
"@push.rocks/smartarray": "^1.1.0",
|
||||
"@push.rocks/smartfile": "^11.2.0",
|
||||
"@push.rocks/smartpath": "^5.0.18",
|
||||
"@push.rocks/smartpdf": "^3.1.5",
|
||||
"@push.rocks/smartpromise": "^4.0.3",
|
||||
"@push.rocks/smartrequest": "^2.0.22",
|
||||
"@push.rocks/webstream": "^1.0.8",
|
||||
"openai": "^4.38.5"
|
||||
"@push.rocks/smartpdf": "^3.2.2",
|
||||
"@push.rocks/smartpromise": "^4.2.3",
|
||||
"@push.rocks/smartrequest": "^2.0.23",
|
||||
"@push.rocks/webstream": "^1.0.10",
|
||||
"openai": "^4.85.4"
|
||||
},
|
||||
"repository": {
|
||||
"type": "git",
|
||||
"url": "git+https://code.foss.global/push.rocks/smartai.git"
|
||||
"url": "https://code.foss.global/push.rocks/smartai.git"
|
||||
},
|
||||
"bugs": {
|
||||
"url": "https://code.foss.global/push.rocks/smartai/issues"
|
||||
},
|
||||
"homepage": "https://code.foss.global/push.rocks/smartai#readme",
|
||||
"homepage": "https://code.foss.global/push.rocks/smartai",
|
||||
"browserslist": [
|
||||
"last 1 chrome versions"
|
||||
],
|
||||
@ -66,5 +66,10 @@
|
||||
"audio responses",
|
||||
"text-to-speech",
|
||||
"streaming chat"
|
||||
],
|
||||
"pnpm": {
|
||||
"onlyBuiltDependencies": [
|
||||
"puppeteer"
|
||||
]
|
||||
}
|
||||
}
|
||||
|
||||
11468
pnpm-lock.yaml
generated
11468
pnpm-lock.yaml
generated
File diff suppressed because it is too large
Load Diff
332
readme.md
332
readme.md
@ -1,95 +1,329 @@
|
||||
# @push.rocks/smartai
|
||||
|
||||
Provides a standardized interface for integrating and conversing with multiple AI models, supporting operations like chat and potentially audio responses.
|
||||
[](https://www.npmjs.com/package/@push.rocks/smartai)
|
||||
|
||||
## Install
|
||||
SmartAi is a comprehensive TypeScript library that provides a standardized interface for integrating and interacting with multiple AI models. It supports a range of operations from synchronous and streaming chat to audio generation, document processing, and vision tasks.
|
||||
|
||||
To add @push.rocks/smartai to your project, run the following command in your terminal:
|
||||
## Table of Contents
|
||||
|
||||
- [Features](#features)
|
||||
- [Installation](#installation)
|
||||
- [Supported AI Providers](#supported-ai-providers)
|
||||
- [Quick Start](#quick-start)
|
||||
- [Usage Examples](#usage-examples)
|
||||
- [Chat Interactions](#chat-interactions)
|
||||
- [Streaming Chat](#streaming-chat)
|
||||
- [Audio Generation](#audio-generation)
|
||||
- [Document Processing](#document-processing)
|
||||
- [Vision Processing](#vision-processing)
|
||||
- [Error Handling](#error-handling)
|
||||
- [Development](#development)
|
||||
- [Running Tests](#running-tests)
|
||||
- [Building the Project](#building-the-project)
|
||||
- [Contributing](#contributing)
|
||||
- [License](#license)
|
||||
- [Legal Information](#legal-information)
|
||||
|
||||
## Features
|
||||
|
||||
- **Unified API:** Seamlessly integrate multiple AI providers with a consistent interface.
|
||||
- **Chat & Streaming:** Support for both synchronous and real-time streaming chat interactions.
|
||||
- **Audio & Vision:** Generate audio responses and perform detailed image analysis.
|
||||
- **Document Processing:** Analyze PDFs and other documents using vision models.
|
||||
- **Extensible:** Easily extend the library to support additional AI providers.
|
||||
|
||||
## Installation
|
||||
|
||||
To install SmartAi, run the following command:
|
||||
|
||||
```bash
|
||||
npm install @push.rocks/smartai
|
||||
```
|
||||
|
||||
This command installs the package and adds it to your project's dependencies.
|
||||
This will add the package to your project’s dependencies.
|
||||
|
||||
## Usage
|
||||
## Supported AI Providers
|
||||
|
||||
The `@push.rocks/smartai` package is a comprehensive solution for integrating and interacting with various AI models, designed to support operations ranging from chat interactions to possibly handling audio responses. This documentation will guide you through the process of utilizing `@push.rocks/smartai` in your applications, focusing on TypeScript and ESM syntax to demonstrate its full capabilities.
|
||||
SmartAi supports multiple AI providers. Configure each provider with its corresponding token or settings:
|
||||
|
||||
### Getting Started
|
||||
### OpenAI
|
||||
|
||||
Before you begin, ensure you have installed the package in your project as described in the **Install** section above. Once installed, you can start integrating AI functionalities into your application.
|
||||
- **Models:** GPT-4, GPT-3.5-turbo, GPT-4-vision-preview
|
||||
- **Features:** Chat, Streaming, Audio Generation, Vision, Document Processing
|
||||
- **Configuration Example:**
|
||||
|
||||
### Initializing SmartAi
|
||||
```typescript
|
||||
openaiToken: 'your-openai-token'
|
||||
```
|
||||
|
||||
The first step is to import and initialize the `SmartAi` class with appropriate options, including tokens for the AI services you plan to use:
|
||||
### X.AI
|
||||
|
||||
- **Models:** Grok-2-latest
|
||||
- **Features:** Chat, Streaming, Document Processing
|
||||
- **Configuration Example:**
|
||||
|
||||
```typescript
|
||||
xaiToken: 'your-xai-token'
|
||||
```
|
||||
|
||||
### Anthropic
|
||||
|
||||
- **Models:** Claude-3-opus-20240229
|
||||
- **Features:** Chat, Streaming, Vision, Document Processing
|
||||
- **Configuration Example:**
|
||||
|
||||
```typescript
|
||||
anthropicToken: 'your-anthropic-token'
|
||||
```
|
||||
|
||||
### Perplexity
|
||||
|
||||
- **Models:** Mixtral-8x7b-instruct
|
||||
- **Features:** Chat, Streaming
|
||||
- **Configuration Example:**
|
||||
|
||||
```typescript
|
||||
perplexityToken: 'your-perplexity-token'
|
||||
```
|
||||
|
||||
### Groq
|
||||
|
||||
- **Models:** Llama-3.3-70b-versatile
|
||||
- **Features:** Chat, Streaming
|
||||
- **Configuration Example:**
|
||||
|
||||
```typescript
|
||||
groqToken: 'your-groq-token'
|
||||
```
|
||||
|
||||
### Ollama
|
||||
|
||||
- **Models:** Configurable (default: llama2; use llava for vision/document tasks)
|
||||
- **Features:** Chat, Streaming, Vision, Document Processing
|
||||
- **Configuration Example:**
|
||||
|
||||
```typescript
|
||||
ollama: {
|
||||
baseUrl: 'http://localhost:11434', // Optional
|
||||
model: 'llama2', // Optional
|
||||
visionModel: 'llava' // Optional for vision and document tasks
|
||||
}
|
||||
```
|
||||
|
||||
### Exo
|
||||
|
||||
- **Models:** Configurable (supports LLaMA, Mistral, LlaVA, Qwen, and Deepseek)
|
||||
- **Features:** Chat, Streaming
|
||||
- **Configuration Example:**
|
||||
|
||||
```typescript
|
||||
exo: {
|
||||
baseUrl: 'http://localhost:8080/v1', // Optional
|
||||
apiKey: 'your-api-key' // Optional for local deployments
|
||||
}
|
||||
```
|
||||
|
||||
## Quick Start
|
||||
|
||||
Initialize SmartAi with the provider configurations you plan to use:
|
||||
|
||||
```typescript
|
||||
import { SmartAi } from '@push.rocks/smartai';
|
||||
|
||||
const smartAi = new SmartAi({
|
||||
openaiToken: 'your-openai-access-token',
|
||||
anthropicToken: 'your-anthropic-access-token'
|
||||
openaiToken: 'your-openai-token',
|
||||
xaiToken: 'your-xai-token',
|
||||
anthropicToken: 'your-anthropic-token',
|
||||
perplexityToken: 'your-perplexity-token',
|
||||
groqToken: 'your-groq-token',
|
||||
ollama: {
|
||||
baseUrl: 'http://localhost:11434',
|
||||
model: 'llama2'
|
||||
},
|
||||
exo: {
|
||||
baseUrl: 'http://localhost:8080/v1',
|
||||
apiKey: 'your-api-key'
|
||||
}
|
||||
});
|
||||
|
||||
await smartAi.start();
|
||||
```
|
||||
|
||||
### Creating Conversations with AI
|
||||
## Usage Examples
|
||||
|
||||
`SmartAi` provides a flexible interface to create and manage conversations with different AI providers. You can create a conversation with any supported AI provider like OpenAI or Anthropic by specifying the provider you want to use:
|
||||
### Chat Interactions
|
||||
|
||||
**Synchronous Chat:**
|
||||
|
||||
```typescript
|
||||
const openAiConversation = await smartAi.createConversation('openai');
|
||||
const anthropicConversation = await smartAi.createConversation('anthropic');
|
||||
```
|
||||
|
||||
### Chatting with AI
|
||||
|
||||
Once you have a conversation instance, you can start sending messages to the AI and receive responses. Each conversation object provides methods to interact in a synchronous or asynchronous manner, depending on your use case.
|
||||
|
||||
#### Synchronous Chat Example
|
||||
|
||||
Here's how you can have a synchronous chat with OpenAI:
|
||||
|
||||
```typescript
|
||||
const response = await openAiConversation.chat({
|
||||
systemMessage: 'This is a greeting from the system.',
|
||||
userMessage: 'Hello, AI! How are you today?',
|
||||
messageHistory: [] // Previous messages in the conversation
|
||||
const response = await smartAi.openaiProvider.chat({
|
||||
systemMessage: 'You are a helpful assistant.',
|
||||
userMessage: 'What is the capital of France?',
|
||||
messageHistory: [] // Include previous conversation messages if applicable
|
||||
});
|
||||
|
||||
console.log(response.message); // Log the response from AI
|
||||
console.log(response.message);
|
||||
```
|
||||
|
||||
#### Streaming Chat Example
|
||||
### Streaming Chat
|
||||
|
||||
For real-time, streaming interactions, you can utilize the streaming capabilities provided by the conversation object. This enables a continuous exchange of messages between your application and the AI:
|
||||
**Real-Time Streaming:**
|
||||
|
||||
```typescript
|
||||
const inputStreamWriter = openAiConversation.getInputStreamWriter();
|
||||
const outputStream = openAiConversation.getOutputStream();
|
||||
const textEncoder = new TextEncoder();
|
||||
const textDecoder = new TextDecoder();
|
||||
|
||||
inputStreamWriter.write('Hello, AI! Can you stream responses?');
|
||||
// Create a transform stream for sending and receiving data
|
||||
const { writable, readable } = new TransformStream();
|
||||
const writer = writable.getWriter();
|
||||
|
||||
const reader = outputStream.getReader();
|
||||
reader.read().then(function processText({done, value}) {
|
||||
if (done) {
|
||||
console.log('Stream finished.');
|
||||
return;
|
||||
}
|
||||
console.log('AI says:', value);
|
||||
reader.read().then(processText); // Continue reading messages
|
||||
const message = {
|
||||
role: 'user',
|
||||
content: 'Tell me a story about a brave knight'
|
||||
};
|
||||
|
||||
writer.write(textEncoder.encode(JSON.stringify(message) + '\n'));
|
||||
|
||||
// Start streaming the response
|
||||
const stream = await smartAi.openaiProvider.chatStream(readable);
|
||||
const reader = stream.getReader();
|
||||
|
||||
while (true) {
|
||||
const { done, value } = await reader.read();
|
||||
if (done) break;
|
||||
console.log('AI:', value);
|
||||
}
|
||||
```
|
||||
|
||||
### Audio Generation
|
||||
|
||||
Generate audio (supported by providers like OpenAI):
|
||||
|
||||
```typescript
|
||||
const audioStream = await smartAi.openaiProvider.audio({
|
||||
message: 'Hello, this is a test of text-to-speech'
|
||||
});
|
||||
|
||||
// Process the audio stream, for example, play it or save to a file.
|
||||
```
|
||||
|
||||
### Document Processing
|
||||
|
||||
Analyze and extract key information from documents:
|
||||
|
||||
```typescript
|
||||
// Example using OpenAI
|
||||
const documentResult = await smartAi.openaiProvider.document({
|
||||
systemMessage: 'Classify the document type',
|
||||
userMessage: 'What type of document is this?',
|
||||
messageHistory: [],
|
||||
pdfDocuments: [pdfBuffer] // Uint8Array containing the PDF content
|
||||
});
|
||||
```
|
||||
|
||||
### Extending Conversations
|
||||
Other providers (e.g., Ollama and Anthropic) follow a similar pattern:
|
||||
|
||||
The modular design of `@push.rocks/smartai` allows you to extend conversations with additional features, such as handling audio responses or integrating other AI-powered functionalities. Utilize the provided AI providers' APIs to explore and implement a wide range of AI interactions within your conversations.
|
||||
```typescript
|
||||
// Using Ollama for document processing
|
||||
const ollamaResult = await smartAi.ollamaProvider.document({
|
||||
systemMessage: 'You are a document analysis assistant',
|
||||
userMessage: 'Extract key information from this document',
|
||||
messageHistory: [],
|
||||
pdfDocuments: [pdfBuffer]
|
||||
});
|
||||
```
|
||||
|
||||
### Conclusion
|
||||
```typescript
|
||||
// Using Anthropic for document processing
|
||||
const anthropicResult = await smartAi.anthropicProvider.document({
|
||||
systemMessage: 'Analyze the document',
|
||||
userMessage: 'Please extract the main points',
|
||||
messageHistory: [],
|
||||
pdfDocuments: [pdfBuffer]
|
||||
});
|
||||
```
|
||||
|
||||
With `@push.rocks/smartai`, integrating AI functionalities into your applications becomes streamlined and efficient. By leveraging the standardized interface provided by the package, you can easily converse with multiple AI models, expanding the capabilities of your applications with cutting-edge AI features. Whether you're implementing simple chat interactions or complex, real-time communication flows, `@push.rocks/smartai` offers the tools and flexibility needed to create engaging, AI-enhanced experiences.
|
||||
### Vision Processing
|
||||
|
||||
Analyze images with vision capabilities:
|
||||
|
||||
```typescript
|
||||
// Using OpenAI GPT-4 Vision
|
||||
const imageDescription = await smartAi.openaiProvider.vision({
|
||||
image: imageBuffer, // Uint8Array containing image data
|
||||
prompt: 'What do you see in this image?'
|
||||
});
|
||||
|
||||
// Using Ollama for vision tasks
|
||||
const ollamaImageAnalysis = await smartAi.ollamaProvider.vision({
|
||||
image: imageBuffer,
|
||||
prompt: 'Analyze this image in detail'
|
||||
});
|
||||
|
||||
// Using Anthropic for vision analysis
|
||||
const anthropicImageAnalysis = await smartAi.anthropicProvider.vision({
|
||||
image: imageBuffer,
|
||||
prompt: 'Describe the contents of this image'
|
||||
});
|
||||
```
|
||||
|
||||
## Error Handling
|
||||
|
||||
Always wrap API calls in try-catch blocks to manage errors effectively:
|
||||
|
||||
```typescript
|
||||
try {
|
||||
const response = await smartAi.openaiProvider.chat({
|
||||
systemMessage: 'You are a helpful assistant.',
|
||||
userMessage: 'Hello!',
|
||||
messageHistory: []
|
||||
});
|
||||
console.log(response.message);
|
||||
} catch (error: any) {
|
||||
console.error('AI provider error:', error.message);
|
||||
}
|
||||
```
|
||||
|
||||
## Development
|
||||
|
||||
### Running Tests
|
||||
|
||||
To run the test suite, use the following command:
|
||||
|
||||
```bash
|
||||
npm run test
|
||||
```
|
||||
|
||||
Ensure your environment is configured with the appropriate tokens and settings for the providers you are testing.
|
||||
|
||||
### Building the Project
|
||||
|
||||
Compile the TypeScript code and build the package using:
|
||||
|
||||
```bash
|
||||
npm run build
|
||||
```
|
||||
|
||||
This command prepares the library for distribution.
|
||||
|
||||
## Contributing
|
||||
|
||||
Contributions are welcome! Please follow these steps:
|
||||
|
||||
1. Fork the repository.
|
||||
2. Create a feature branch:
|
||||
```bash
|
||||
git checkout -b feature/my-feature
|
||||
```
|
||||
3. Commit your changes with clear messages:
|
||||
```bash
|
||||
git commit -m 'Add new feature'
|
||||
```
|
||||
4. Push your branch to your fork:
|
||||
```bash
|
||||
git push origin feature/my-feature
|
||||
```
|
||||
5. Open a Pull Request with a detailed description of your changes.
|
||||
|
||||
## License and Legal Information
|
||||
|
||||
|
||||
@ -25,14 +25,14 @@ tap.test('should create chat response with openai', async () => {
|
||||
],
|
||||
});
|
||||
console.log(`userMessage: ${userMessage}`);
|
||||
console.log(response.message.content);
|
||||
console.log(response.message);
|
||||
});
|
||||
|
||||
tap.test('should document a pdf', async () => {
|
||||
const pdfUrl = 'https://www.w3.org/WAI/ER/tests/xhtml/testfiles/resources/pdf/dummy.pdf';
|
||||
const pdfResponse = await smartrequest.getBinary(pdfUrl);
|
||||
const result = await testSmartai.openaiProvider.document({
|
||||
systemMessage: 'Classify the document. Only the following answers are allowed: "invoice", "bank account statement", "contract", "other"',
|
||||
systemMessage: 'Classify the document. Only the following answers are allowed: "invoice", "bank account statement", "contract", "other". The answer should only contain the keyword for machine use.',
|
||||
userMessage: "Classify the document.",
|
||||
messageHistory: [],
|
||||
pdfDocuments: [pdfResponse.body],
|
||||
|
||||
@ -1,8 +1,8 @@
|
||||
/**
|
||||
* autocreated commitinfo by @pushrocks/commitinfo
|
||||
* autocreated commitinfo by @push.rocks/commitinfo
|
||||
*/
|
||||
export const commitinfo = {
|
||||
name: '@push.rocks/smartai',
|
||||
version: '0.0.13',
|
||||
version: '0.4.2',
|
||||
description: 'A TypeScript library for integrating and interacting with multiple AI models, offering capabilities for chat and potentially audio responses.'
|
||||
}
|
||||
|
||||
@ -1,32 +1,86 @@
|
||||
/**
|
||||
* Message format for chat interactions
|
||||
*/
|
||||
export interface ChatMessage {
|
||||
role: 'assistant' | 'user' | 'system';
|
||||
content: string;
|
||||
}
|
||||
|
||||
/**
|
||||
* Options for chat interactions
|
||||
*/
|
||||
export interface ChatOptions {
|
||||
systemMessage: string;
|
||||
userMessage: string;
|
||||
messageHistory: ChatMessage[];
|
||||
}
|
||||
|
||||
/**
|
||||
* Response format for chat interactions
|
||||
*/
|
||||
export interface ChatResponse {
|
||||
role: 'assistant';
|
||||
message: string;
|
||||
}
|
||||
|
||||
/**
|
||||
* Abstract base class for multi-modal AI models.
|
||||
* Provides a common interface for different AI providers (OpenAI, Anthropic, Perplexity, Ollama)
|
||||
*/
|
||||
export abstract class MultiModalModel {
|
||||
/**
|
||||
* starts the model
|
||||
* Initializes the model and any necessary resources
|
||||
* Should be called before using any other methods
|
||||
*/
|
||||
abstract start(): Promise<void>;
|
||||
|
||||
/**
|
||||
* stops the model
|
||||
* Cleans up any resources used by the model
|
||||
* Should be called when the model is no longer needed
|
||||
*/
|
||||
abstract stop(): Promise<void>;
|
||||
|
||||
public abstract chat(optionsArg: {
|
||||
systemMessage: string,
|
||||
userMessage: string,
|
||||
messageHistory: {
|
||||
role: 'assistant' | 'user';
|
||||
content: string;
|
||||
}[]
|
||||
}): Promise<{
|
||||
role: 'assistant';
|
||||
message: string;
|
||||
}>
|
||||
/**
|
||||
* Synchronous chat interaction with the model
|
||||
* @param optionsArg Options containing system message, user message, and message history
|
||||
* @returns Promise resolving to the assistant's response
|
||||
*/
|
||||
public abstract chat(optionsArg: ChatOptions): Promise<ChatResponse>;
|
||||
|
||||
/**
|
||||
* Defines a streaming interface for chat interactions.
|
||||
* The implementation will vary based on the specific AI model.
|
||||
* @param input
|
||||
* Streaming interface for chat interactions
|
||||
* Allows for real-time responses from the model
|
||||
* @param input Stream of user messages
|
||||
* @returns Stream of model responses
|
||||
*/
|
||||
public abstract chatStream(input: ReadableStream<string>): Promise<ReadableStream<string>>;
|
||||
public abstract chatStream(input: ReadableStream<Uint8Array>): Promise<ReadableStream<string>>;
|
||||
|
||||
/**
|
||||
* Text-to-speech conversion
|
||||
* @param optionsArg Options containing the message to convert to speech
|
||||
* @returns Promise resolving to a readable stream of audio data
|
||||
* @throws Error if the provider doesn't support audio generation
|
||||
*/
|
||||
public abstract audio(optionsArg: { message: string }): Promise<NodeJS.ReadableStream>;
|
||||
|
||||
/**
|
||||
* Vision-language processing
|
||||
* @param optionsArg Options containing the image and prompt for analysis
|
||||
* @returns Promise resolving to the model's description or analysis of the image
|
||||
* @throws Error if the provider doesn't support vision tasks
|
||||
*/
|
||||
public abstract vision(optionsArg: { image: Buffer; prompt: string }): Promise<string>;
|
||||
|
||||
/**
|
||||
* Document analysis and processing
|
||||
* @param optionsArg Options containing system message, user message, PDF documents, and message history
|
||||
* @returns Promise resolving to the model's analysis of the documents
|
||||
* @throws Error if the provider doesn't support document processing
|
||||
*/
|
||||
public abstract document(optionsArg: {
|
||||
systemMessage: string;
|
||||
userMessage: string;
|
||||
pdfDocuments: Uint8Array[];
|
||||
messageHistory: ChatMessage[];
|
||||
}): Promise<{ message: any }>;
|
||||
}
|
||||
|
||||
@ -48,6 +48,18 @@ export class Conversation {
|
||||
return conversation;
|
||||
}
|
||||
|
||||
public static async createWithExo(smartaiRefArg: SmartAi) {
|
||||
if (!smartaiRefArg.exoProvider) {
|
||||
throw new Error('Exo provider not available');
|
||||
}
|
||||
const conversation = new Conversation(smartaiRefArg, {
|
||||
processFunction: async (input) => {
|
||||
return '' // TODO implement proper streaming
|
||||
}
|
||||
});
|
||||
return conversation;
|
||||
}
|
||||
|
||||
public static async createWithOllama(smartaiRefArg: SmartAi) {
|
||||
if (!smartaiRefArg.ollamaProvider) {
|
||||
throw new Error('Ollama provider not available');
|
||||
@ -60,6 +72,30 @@ export class Conversation {
|
||||
return conversation;
|
||||
}
|
||||
|
||||
public static async createWithGroq(smartaiRefArg: SmartAi) {
|
||||
if (!smartaiRefArg.groqProvider) {
|
||||
throw new Error('Groq provider not available');
|
||||
}
|
||||
const conversation = new Conversation(smartaiRefArg, {
|
||||
processFunction: async (input) => {
|
||||
return '' // TODO implement proper streaming
|
||||
}
|
||||
});
|
||||
return conversation;
|
||||
}
|
||||
|
||||
public static async createWithXai(smartaiRefArg: SmartAi) {
|
||||
if (!smartaiRefArg.xaiProvider) {
|
||||
throw new Error('XAI provider not available');
|
||||
}
|
||||
const conversation = new Conversation(smartaiRefArg, {
|
||||
processFunction: async (input) => {
|
||||
return '' // TODO implement proper streaming
|
||||
}
|
||||
});
|
||||
return conversation;
|
||||
}
|
||||
|
||||
// INSTANCE
|
||||
smartaiRef: SmartAi
|
||||
private systemMessage: string;
|
||||
|
||||
@ -1,18 +1,32 @@
|
||||
import { Conversation } from './classes.conversation.js';
|
||||
import * as plugins from './plugins.js';
|
||||
import { AnthropicProvider } from './provider.anthropic.js';
|
||||
import type { OllamaProvider } from './provider.ollama.js';
|
||||
import { OllamaProvider } from './provider.ollama.js';
|
||||
import { OpenAiProvider } from './provider.openai.js';
|
||||
import type { PerplexityProvider } from './provider.perplexity.js';
|
||||
import { PerplexityProvider } from './provider.perplexity.js';
|
||||
import { ExoProvider } from './provider.exo.js';
|
||||
import { GroqProvider } from './provider.groq.js';
|
||||
import { XAIProvider } from './provider.xai.js';
|
||||
|
||||
|
||||
export interface ISmartAiOptions {
|
||||
openaiToken?: string;
|
||||
anthropicToken?: string;
|
||||
perplexityToken?: string;
|
||||
groqToken?: string;
|
||||
xaiToken?: string;
|
||||
exo?: {
|
||||
baseUrl?: string;
|
||||
apiKey?: string;
|
||||
};
|
||||
ollama?: {
|
||||
baseUrl?: string;
|
||||
model?: string;
|
||||
visionModel?: string;
|
||||
};
|
||||
}
|
||||
|
||||
export type TProvider = 'openai' | 'anthropic' | 'perplexity' | 'ollama';
|
||||
export type TProvider = 'openai' | 'anthropic' | 'perplexity' | 'ollama' | 'exo' | 'groq' | 'xai';
|
||||
|
||||
export class SmartAi {
|
||||
public options: ISmartAiOptions;
|
||||
@ -21,6 +35,9 @@ export class SmartAi {
|
||||
public anthropicProvider: AnthropicProvider;
|
||||
public perplexityProvider: PerplexityProvider;
|
||||
public ollamaProvider: OllamaProvider;
|
||||
public exoProvider: ExoProvider;
|
||||
public groqProvider: GroqProvider;
|
||||
public xaiProvider: XAIProvider;
|
||||
|
||||
constructor(optionsArg: ISmartAiOptions) {
|
||||
this.options = optionsArg;
|
||||
@ -37,6 +54,40 @@ export class SmartAi {
|
||||
this.anthropicProvider = new AnthropicProvider({
|
||||
anthropicToken: this.options.anthropicToken,
|
||||
});
|
||||
await this.anthropicProvider.start();
|
||||
}
|
||||
if (this.options.perplexityToken) {
|
||||
this.perplexityProvider = new PerplexityProvider({
|
||||
perplexityToken: this.options.perplexityToken,
|
||||
});
|
||||
await this.perplexityProvider.start();
|
||||
}
|
||||
if (this.options.groqToken) {
|
||||
this.groqProvider = new GroqProvider({
|
||||
groqToken: this.options.groqToken,
|
||||
});
|
||||
await this.groqProvider.start();
|
||||
}
|
||||
if (this.options.xaiToken) {
|
||||
this.xaiProvider = new XAIProvider({
|
||||
xaiToken: this.options.xaiToken,
|
||||
});
|
||||
await this.xaiProvider.start();
|
||||
}
|
||||
if (this.options.ollama) {
|
||||
this.ollamaProvider = new OllamaProvider({
|
||||
baseUrl: this.options.ollama.baseUrl,
|
||||
model: this.options.ollama.model,
|
||||
visionModel: this.options.ollama.visionModel,
|
||||
});
|
||||
await this.ollamaProvider.start();
|
||||
}
|
||||
if (this.options.exo) {
|
||||
this.exoProvider = new ExoProvider({
|
||||
exoBaseUrl: this.options.exo.baseUrl,
|
||||
apiKey: this.options.exo.apiKey,
|
||||
});
|
||||
await this.exoProvider.start();
|
||||
}
|
||||
}
|
||||
|
||||
@ -47,6 +98,8 @@ export class SmartAi {
|
||||
*/
|
||||
createConversation(provider: TProvider) {
|
||||
switch (provider) {
|
||||
case 'exo':
|
||||
return Conversation.createWithExo(this);
|
||||
case 'openai':
|
||||
return Conversation.createWithOpenAi(this);
|
||||
case 'anthropic':
|
||||
@ -55,6 +108,10 @@ export class SmartAi {
|
||||
return Conversation.createWithPerplexity(this);
|
||||
case 'ollama':
|
||||
return Conversation.createWithOllama(this);
|
||||
case 'groq':
|
||||
return Conversation.createWithGroq(this);
|
||||
case 'xai':
|
||||
return Conversation.createWithXai(this);
|
||||
default:
|
||||
throw new Error('Provider not available');
|
||||
}
|
||||
|
||||
@ -1,6 +1,10 @@
|
||||
import * as plugins from './plugins.js';
|
||||
import * as paths from './paths.js';
|
||||
import { MultiModalModel } from './abstract.classes.multimodal.js';
|
||||
import type { ChatOptions, ChatResponse, ChatMessage } from './abstract.classes.multimodal.js';
|
||||
import type { ImageBlockParam, TextBlockParam } from '@anthropic-ai/sdk/resources/messages';
|
||||
|
||||
type ContentBlock = ImageBlockParam | TextBlockParam;
|
||||
|
||||
export interface IAnthropicProviderOptions {
|
||||
anthropicToken: string;
|
||||
@ -23,40 +27,214 @@ export class AnthropicProvider extends MultiModalModel {
|
||||
|
||||
async stop() {}
|
||||
|
||||
public async chatStream(input: ReadableStream<string>): Promise<ReadableStream<string>> {
|
||||
// TODO: implement for OpenAI
|
||||
public async chatStream(input: ReadableStream<Uint8Array>): Promise<ReadableStream<string>> {
|
||||
// Create a TextDecoder to handle incoming chunks
|
||||
const decoder = new TextDecoder();
|
||||
let buffer = '';
|
||||
let currentMessage: { role: string; content: string; } | null = null;
|
||||
|
||||
const returnStream = new ReadableStream();
|
||||
return returnStream;
|
||||
// Create a TransformStream to process the input
|
||||
const transform = new TransformStream<Uint8Array, string>({
|
||||
async transform(chunk, controller) {
|
||||
buffer += decoder.decode(chunk, { stream: true });
|
||||
|
||||
// Try to parse complete JSON messages from the buffer
|
||||
while (true) {
|
||||
const newlineIndex = buffer.indexOf('\n');
|
||||
if (newlineIndex === -1) break;
|
||||
|
||||
const line = buffer.slice(0, newlineIndex);
|
||||
buffer = buffer.slice(newlineIndex + 1);
|
||||
|
||||
if (line.trim()) {
|
||||
try {
|
||||
const message = JSON.parse(line);
|
||||
currentMessage = {
|
||||
role: message.role || 'user',
|
||||
content: message.content || '',
|
||||
};
|
||||
} catch (e) {
|
||||
console.error('Failed to parse message:', e);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// If we have a complete message, send it to Anthropic
|
||||
if (currentMessage) {
|
||||
const stream = await this.anthropicApiClient.messages.create({
|
||||
model: 'claude-3-opus-20240229',
|
||||
messages: [{ role: currentMessage.role, content: currentMessage.content }],
|
||||
system: '',
|
||||
stream: true,
|
||||
max_tokens: 4000,
|
||||
});
|
||||
|
||||
// Process each chunk from Anthropic
|
||||
for await (const chunk of stream) {
|
||||
const content = chunk.delta?.text;
|
||||
if (content) {
|
||||
controller.enqueue(content);
|
||||
}
|
||||
}
|
||||
|
||||
currentMessage = null;
|
||||
}
|
||||
},
|
||||
|
||||
flush(controller) {
|
||||
if (buffer) {
|
||||
try {
|
||||
const message = JSON.parse(buffer);
|
||||
controller.enqueue(message.content || '');
|
||||
} catch (e) {
|
||||
console.error('Failed to parse remaining buffer:', e);
|
||||
}
|
||||
}
|
||||
}
|
||||
});
|
||||
|
||||
// Connect the input to our transform stream
|
||||
return input.pipeThrough(transform);
|
||||
}
|
||||
|
||||
// Implementing the synchronous chat interaction
|
||||
public async chat(optionsArg: {
|
||||
systemMessage: string;
|
||||
userMessage: string;
|
||||
messageHistory: {
|
||||
role: 'assistant' | 'user';
|
||||
content: string;
|
||||
}[];
|
||||
}) {
|
||||
public async chat(optionsArg: ChatOptions): Promise<ChatResponse> {
|
||||
// Convert message history to Anthropic format
|
||||
const messages = optionsArg.messageHistory.map(msg => ({
|
||||
role: msg.role === 'assistant' ? 'assistant' as const : 'user' as const,
|
||||
content: msg.content
|
||||
}));
|
||||
|
||||
const result = await this.anthropicApiClient.messages.create({
|
||||
model: 'claude-3-opus-20240229',
|
||||
system: optionsArg.systemMessage,
|
||||
messages: [
|
||||
...optionsArg.messageHistory,
|
||||
{ role: 'user', content: optionsArg.userMessage },
|
||||
...messages,
|
||||
{ role: 'user' as const, content: optionsArg.userMessage }
|
||||
],
|
||||
max_tokens: 4000,
|
||||
});
|
||||
|
||||
// Extract text content from the response
|
||||
let message = '';
|
||||
for (const block of result.content) {
|
||||
if ('text' in block) {
|
||||
message += block.text;
|
||||
}
|
||||
}
|
||||
|
||||
return {
|
||||
role: result.role as 'assistant',
|
||||
message: result.content.join('\n'),
|
||||
role: 'assistant' as const,
|
||||
message,
|
||||
};
|
||||
}
|
||||
|
||||
private async audio(messageArg: string) {
|
||||
public async audio(optionsArg: { message: string }): Promise<NodeJS.ReadableStream> {
|
||||
// Anthropic does not provide an audio API, so this method is not implemented.
|
||||
throw new Error('Audio generation is not yet supported by Anthropic.');
|
||||
}
|
||||
|
||||
public async vision(optionsArg: { image: Buffer; prompt: string }): Promise<string> {
|
||||
const base64Image = optionsArg.image.toString('base64');
|
||||
|
||||
const content: ContentBlock[] = [
|
||||
{
|
||||
type: 'text',
|
||||
text: optionsArg.prompt
|
||||
},
|
||||
{
|
||||
type: 'image',
|
||||
source: {
|
||||
type: 'base64',
|
||||
media_type: 'image/jpeg',
|
||||
data: base64Image
|
||||
}
|
||||
}
|
||||
];
|
||||
|
||||
const result = await this.anthropicApiClient.messages.create({
|
||||
model: 'claude-3-opus-20240229',
|
||||
messages: [{
|
||||
role: 'user',
|
||||
content
|
||||
}],
|
||||
max_tokens: 1024
|
||||
});
|
||||
|
||||
// Extract text content from the response
|
||||
let message = '';
|
||||
for (const block of result.content) {
|
||||
if ('text' in block) {
|
||||
message += block.text;
|
||||
}
|
||||
}
|
||||
return message;
|
||||
}
|
||||
|
||||
public async document(optionsArg: {
|
||||
systemMessage: string;
|
||||
userMessage: string;
|
||||
pdfDocuments: Uint8Array[];
|
||||
messageHistory: ChatMessage[];
|
||||
}): Promise<{ message: any }> {
|
||||
// Convert PDF documents to images using SmartPDF
|
||||
const smartpdfInstance = new plugins.smartpdf.SmartPdf();
|
||||
let documentImageBytesArray: Uint8Array[] = [];
|
||||
|
||||
for (const pdfDocument of optionsArg.pdfDocuments) {
|
||||
const documentImageArray = await smartpdfInstance.convertPDFToPngBytes(pdfDocument);
|
||||
documentImageBytesArray = documentImageBytesArray.concat(documentImageArray);
|
||||
}
|
||||
|
||||
// Convert message history to Anthropic format
|
||||
const messages = optionsArg.messageHistory.map(msg => ({
|
||||
role: msg.role === 'assistant' ? 'assistant' as const : 'user' as const,
|
||||
content: msg.content
|
||||
}));
|
||||
|
||||
// Create content array with text and images
|
||||
const content: ContentBlock[] = [
|
||||
{
|
||||
type: 'text',
|
||||
text: optionsArg.userMessage
|
||||
}
|
||||
];
|
||||
|
||||
// Add each document page as an image
|
||||
for (const imageBytes of documentImageBytesArray) {
|
||||
content.push({
|
||||
type: 'image',
|
||||
source: {
|
||||
type: 'base64',
|
||||
media_type: 'image/jpeg',
|
||||
data: Buffer.from(imageBytes).toString('base64')
|
||||
}
|
||||
});
|
||||
}
|
||||
|
||||
const result = await this.anthropicApiClient.messages.create({
|
||||
model: 'claude-3-opus-20240229',
|
||||
system: optionsArg.systemMessage,
|
||||
messages: [
|
||||
...messages,
|
||||
{ role: 'user', content }
|
||||
],
|
||||
max_tokens: 4096
|
||||
});
|
||||
|
||||
// Extract text content from the response
|
||||
let message = '';
|
||||
for (const block of result.content) {
|
||||
if ('text' in block) {
|
||||
message += block.text;
|
||||
}
|
||||
}
|
||||
|
||||
return {
|
||||
message: {
|
||||
role: 'assistant',
|
||||
content: message
|
||||
}
|
||||
};
|
||||
}
|
||||
}
|
||||
128
ts/provider.exo.ts
Normal file
128
ts/provider.exo.ts
Normal file
@ -0,0 +1,128 @@
|
||||
import * as plugins from './plugins.js';
|
||||
import * as paths from './paths.js';
|
||||
import { MultiModalModel } from './abstract.classes.multimodal.js';
|
||||
import type { ChatOptions, ChatResponse, ChatMessage } from './abstract.classes.multimodal.js';
|
||||
import type { ChatCompletionMessageParam } from 'openai/resources/chat/completions';
|
||||
|
||||
export interface IExoProviderOptions {
|
||||
exoBaseUrl?: string;
|
||||
apiKey?: string;
|
||||
}
|
||||
|
||||
export class ExoProvider extends MultiModalModel {
|
||||
private options: IExoProviderOptions;
|
||||
public openAiApiClient: plugins.openai.default;
|
||||
|
||||
constructor(optionsArg: IExoProviderOptions = {}) {
|
||||
super();
|
||||
this.options = {
|
||||
exoBaseUrl: 'http://localhost:8080/v1', // Default Exo API endpoint
|
||||
...optionsArg
|
||||
};
|
||||
}
|
||||
|
||||
public async start() {
|
||||
this.openAiApiClient = new plugins.openai.default({
|
||||
apiKey: this.options.apiKey || 'not-needed', // Exo might not require an API key for local deployment
|
||||
baseURL: this.options.exoBaseUrl,
|
||||
});
|
||||
}
|
||||
|
||||
public async stop() {}
|
||||
|
||||
public async chatStream(input: ReadableStream<Uint8Array>): Promise<ReadableStream<string>> {
|
||||
// Create a TextDecoder to handle incoming chunks
|
||||
const decoder = new TextDecoder();
|
||||
let buffer = '';
|
||||
let currentMessage: { role: string; content: string; } | null = null;
|
||||
|
||||
// Create a TransformStream to process the input
|
||||
const transform = new TransformStream<Uint8Array, string>({
|
||||
transform: async (chunk, controller) => {
|
||||
buffer += decoder.decode(chunk, { stream: true });
|
||||
|
||||
// Try to parse complete JSON messages from the buffer
|
||||
while (true) {
|
||||
const newlineIndex = buffer.indexOf('\n');
|
||||
if (newlineIndex === -1) break;
|
||||
|
||||
const line = buffer.slice(0, newlineIndex);
|
||||
buffer = buffer.slice(newlineIndex + 1);
|
||||
|
||||
if (line.trim()) {
|
||||
try {
|
||||
const message = JSON.parse(line);
|
||||
currentMessage = message;
|
||||
|
||||
// Process the message based on its type
|
||||
if (message.type === 'message') {
|
||||
const response = await this.chat({
|
||||
systemMessage: '',
|
||||
userMessage: message.content,
|
||||
messageHistory: [{ role: message.role as 'user' | 'assistant' | 'system', content: message.content }]
|
||||
});
|
||||
|
||||
controller.enqueue(JSON.stringify(response) + '\n');
|
||||
}
|
||||
} catch (error) {
|
||||
console.error('Error processing message:', error);
|
||||
}
|
||||
}
|
||||
}
|
||||
},
|
||||
flush(controller) {
|
||||
if (buffer) {
|
||||
try {
|
||||
const message = JSON.parse(buffer);
|
||||
currentMessage = message;
|
||||
} catch (error) {
|
||||
console.error('Error processing remaining buffer:', error);
|
||||
}
|
||||
}
|
||||
}
|
||||
});
|
||||
|
||||
return input.pipeThrough(transform);
|
||||
}
|
||||
|
||||
public async chat(options: ChatOptions): Promise<ChatResponse> {
|
||||
const messages: ChatCompletionMessageParam[] = [
|
||||
{ role: 'system', content: options.systemMessage },
|
||||
...options.messageHistory,
|
||||
{ role: 'user', content: options.userMessage }
|
||||
];
|
||||
|
||||
try {
|
||||
const response = await this.openAiApiClient.chat.completions.create({
|
||||
model: 'local-model', // Exo uses local models
|
||||
messages: messages,
|
||||
stream: false
|
||||
});
|
||||
|
||||
return {
|
||||
role: 'assistant',
|
||||
message: response.choices[0]?.message?.content || ''
|
||||
};
|
||||
} catch (error) {
|
||||
console.error('Error in chat completion:', error);
|
||||
throw error;
|
||||
}
|
||||
}
|
||||
|
||||
public async audio(optionsArg: { message: string }): Promise<NodeJS.ReadableStream> {
|
||||
throw new Error('Audio generation is not supported by Exo provider');
|
||||
}
|
||||
|
||||
public async vision(optionsArg: { image: Buffer; prompt: string }): Promise<string> {
|
||||
throw new Error('Vision processing is not supported by Exo provider');
|
||||
}
|
||||
|
||||
public async document(optionsArg: {
|
||||
systemMessage: string;
|
||||
userMessage: string;
|
||||
pdfDocuments: Uint8Array[];
|
||||
messageHistory: ChatMessage[];
|
||||
}): Promise<{ message: any }> {
|
||||
throw new Error('Document processing is not supported by Exo provider');
|
||||
}
|
||||
}
|
||||
192
ts/provider.groq.ts
Normal file
192
ts/provider.groq.ts
Normal file
@ -0,0 +1,192 @@
|
||||
import * as plugins from './plugins.js';
|
||||
import * as paths from './paths.js';
|
||||
import { MultiModalModel } from './abstract.classes.multimodal.js';
|
||||
import type { ChatOptions, ChatResponse, ChatMessage } from './abstract.classes.multimodal.js';
|
||||
|
||||
export interface IGroqProviderOptions {
|
||||
groqToken: string;
|
||||
model?: string;
|
||||
}
|
||||
|
||||
export class GroqProvider extends MultiModalModel {
|
||||
private options: IGroqProviderOptions;
|
||||
private baseUrl = 'https://api.groq.com/v1';
|
||||
|
||||
constructor(optionsArg: IGroqProviderOptions) {
|
||||
super();
|
||||
this.options = {
|
||||
...optionsArg,
|
||||
model: optionsArg.model || 'llama-3.3-70b-versatile', // Default model
|
||||
};
|
||||
}
|
||||
|
||||
async start() {}
|
||||
|
||||
async stop() {}
|
||||
|
||||
public async chatStream(input: ReadableStream<Uint8Array>): Promise<ReadableStream<string>> {
|
||||
// Create a TextDecoder to handle incoming chunks
|
||||
const decoder = new TextDecoder();
|
||||
let buffer = '';
|
||||
let currentMessage: { role: string; content: string; } | null = null;
|
||||
|
||||
// Create a TransformStream to process the input
|
||||
const transform = new TransformStream<Uint8Array, string>({
|
||||
transform: async (chunk, controller) => {
|
||||
buffer += decoder.decode(chunk, { stream: true });
|
||||
|
||||
// Try to parse complete JSON messages from the buffer
|
||||
while (true) {
|
||||
const newlineIndex = buffer.indexOf('\n');
|
||||
if (newlineIndex === -1) break;
|
||||
|
||||
const line = buffer.slice(0, newlineIndex);
|
||||
buffer = buffer.slice(newlineIndex + 1);
|
||||
|
||||
if (line.trim()) {
|
||||
try {
|
||||
const message = JSON.parse(line);
|
||||
currentMessage = {
|
||||
role: message.role || 'user',
|
||||
content: message.content || '',
|
||||
};
|
||||
} catch (e) {
|
||||
console.error('Failed to parse message:', e);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// If we have a complete message, send it to Groq
|
||||
if (currentMessage) {
|
||||
const response = await fetch(`${this.baseUrl}/chat/completions`, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Authorization': `Bearer ${this.options.groqToken}`,
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
body: JSON.stringify({
|
||||
model: this.options.model,
|
||||
messages: [{ role: currentMessage.role, content: currentMessage.content }],
|
||||
stream: true,
|
||||
}),
|
||||
});
|
||||
|
||||
// Process each chunk from Groq
|
||||
const reader = response.body?.getReader();
|
||||
if (reader) {
|
||||
try {
|
||||
while (true) {
|
||||
const { done, value } = await reader.read();
|
||||
if (done) break;
|
||||
|
||||
const chunk = new TextDecoder().decode(value);
|
||||
const lines = chunk.split('\n');
|
||||
|
||||
for (const line of lines) {
|
||||
if (line.startsWith('data: ')) {
|
||||
const data = line.slice(6);
|
||||
if (data === '[DONE]') break;
|
||||
|
||||
try {
|
||||
const parsed = JSON.parse(data);
|
||||
const content = parsed.choices[0]?.delta?.content;
|
||||
if (content) {
|
||||
controller.enqueue(content);
|
||||
}
|
||||
} catch (e) {
|
||||
console.error('Failed to parse SSE data:', e);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
} finally {
|
||||
reader.releaseLock();
|
||||
}
|
||||
}
|
||||
|
||||
currentMessage = null;
|
||||
}
|
||||
},
|
||||
|
||||
flush(controller) {
|
||||
if (buffer) {
|
||||
try {
|
||||
const message = JSON.parse(buffer);
|
||||
controller.enqueue(message.content || '');
|
||||
} catch (e) {
|
||||
console.error('Failed to parse remaining buffer:', e);
|
||||
}
|
||||
}
|
||||
}
|
||||
});
|
||||
|
||||
// Connect the input to our transform stream
|
||||
return input.pipeThrough(transform);
|
||||
}
|
||||
|
||||
// Implementing the synchronous chat interaction
|
||||
public async chat(optionsArg: ChatOptions): Promise<ChatResponse> {
|
||||
const messages = [
|
||||
// System message
|
||||
{
|
||||
role: 'system',
|
||||
content: optionsArg.systemMessage,
|
||||
},
|
||||
// Message history
|
||||
...optionsArg.messageHistory.map(msg => ({
|
||||
role: msg.role,
|
||||
content: msg.content,
|
||||
})),
|
||||
// User message
|
||||
{
|
||||
role: 'user',
|
||||
content: optionsArg.userMessage,
|
||||
},
|
||||
];
|
||||
|
||||
const response = await fetch(`${this.baseUrl}/chat/completions`, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Authorization': `Bearer ${this.options.groqToken}`,
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
body: JSON.stringify({
|
||||
model: this.options.model,
|
||||
messages,
|
||||
temperature: 0.7,
|
||||
max_completion_tokens: 1024,
|
||||
stream: false,
|
||||
}),
|
||||
});
|
||||
|
||||
if (!response.ok) {
|
||||
const error = await response.json();
|
||||
throw new Error(`Groq API error: ${error.message || response.statusText}`);
|
||||
}
|
||||
|
||||
const result = await response.json();
|
||||
|
||||
return {
|
||||
role: 'assistant',
|
||||
message: result.choices[0].message.content,
|
||||
};
|
||||
}
|
||||
|
||||
public async audio(optionsArg: { message: string }): Promise<NodeJS.ReadableStream> {
|
||||
// Groq does not provide an audio API, so this method is not implemented.
|
||||
throw new Error('Audio generation is not yet supported by Groq.');
|
||||
}
|
||||
|
||||
public async vision(optionsArg: { image: Buffer; prompt: string }): Promise<string> {
|
||||
throw new Error('Vision tasks are not yet supported by Groq.');
|
||||
}
|
||||
|
||||
public async document(optionsArg: {
|
||||
systemMessage: string;
|
||||
userMessage: string;
|
||||
pdfDocuments: Uint8Array[];
|
||||
messageHistory: ChatMessage[];
|
||||
}): Promise<{ message: any }> {
|
||||
throw new Error('Document processing is not yet supported by Groq.');
|
||||
}
|
||||
}
|
||||
@ -1,3 +1,252 @@
|
||||
import * as plugins from './plugins.js';
|
||||
import * as paths from './paths.js';
|
||||
import { MultiModalModel } from './abstract.classes.multimodal.js';
|
||||
import type { ChatOptions, ChatResponse, ChatMessage } from './abstract.classes.multimodal.js';
|
||||
|
||||
export class OllamaProvider {}
|
||||
export interface IOllamaProviderOptions {
|
||||
baseUrl?: string;
|
||||
model?: string;
|
||||
visionModel?: string; // Model to use for vision tasks (e.g. 'llava')
|
||||
}
|
||||
|
||||
export class OllamaProvider extends MultiModalModel {
|
||||
private options: IOllamaProviderOptions;
|
||||
private baseUrl: string;
|
||||
private model: string;
|
||||
private visionModel: string;
|
||||
|
||||
constructor(optionsArg: IOllamaProviderOptions = {}) {
|
||||
super();
|
||||
this.options = optionsArg;
|
||||
this.baseUrl = optionsArg.baseUrl || 'http://localhost:11434';
|
||||
this.model = optionsArg.model || 'llama2';
|
||||
this.visionModel = optionsArg.visionModel || 'llava';
|
||||
}
|
||||
|
||||
async start() {
|
||||
// Verify Ollama is running
|
||||
try {
|
||||
const response = await fetch(`${this.baseUrl}/api/tags`);
|
||||
if (!response.ok) {
|
||||
throw new Error('Failed to connect to Ollama server');
|
||||
}
|
||||
} catch (error) {
|
||||
throw new Error(`Failed to connect to Ollama server at ${this.baseUrl}: ${error.message}`);
|
||||
}
|
||||
}
|
||||
|
||||
async stop() {}
|
||||
|
||||
public async chatStream(input: ReadableStream<Uint8Array>): Promise<ReadableStream<string>> {
|
||||
// Create a TextDecoder to handle incoming chunks
|
||||
const decoder = new TextDecoder();
|
||||
let buffer = '';
|
||||
let currentMessage: { role: string; content: string; } | null = null;
|
||||
|
||||
// Create a TransformStream to process the input
|
||||
const transform = new TransformStream<Uint8Array, string>({
|
||||
transform: async (chunk, controller) => {
|
||||
buffer += decoder.decode(chunk, { stream: true });
|
||||
|
||||
// Try to parse complete JSON messages from the buffer
|
||||
while (true) {
|
||||
const newlineIndex = buffer.indexOf('\n');
|
||||
if (newlineIndex === -1) break;
|
||||
|
||||
const line = buffer.slice(0, newlineIndex);
|
||||
buffer = buffer.slice(newlineIndex + 1);
|
||||
|
||||
if (line.trim()) {
|
||||
try {
|
||||
const message = JSON.parse(line);
|
||||
currentMessage = {
|
||||
role: message.role || 'user',
|
||||
content: message.content || '',
|
||||
};
|
||||
} catch (e) {
|
||||
console.error('Failed to parse message:', e);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// If we have a complete message, send it to Ollama
|
||||
if (currentMessage) {
|
||||
const response = await fetch(`${this.baseUrl}/api/chat`, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
body: JSON.stringify({
|
||||
model: this.model,
|
||||
messages: [{ role: currentMessage.role, content: currentMessage.content }],
|
||||
stream: true,
|
||||
}),
|
||||
});
|
||||
|
||||
// Process each chunk from Ollama
|
||||
const reader = response.body?.getReader();
|
||||
if (reader) {
|
||||
try {
|
||||
while (true) {
|
||||
const { done, value } = await reader.read();
|
||||
if (done) break;
|
||||
|
||||
const chunk = new TextDecoder().decode(value);
|
||||
const lines = chunk.split('\n');
|
||||
|
||||
for (const line of lines) {
|
||||
if (line.trim()) {
|
||||
try {
|
||||
const parsed = JSON.parse(line);
|
||||
const content = parsed.message?.content;
|
||||
if (content) {
|
||||
controller.enqueue(content);
|
||||
}
|
||||
} catch (e) {
|
||||
console.error('Failed to parse Ollama response:', e);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
} finally {
|
||||
reader.releaseLock();
|
||||
}
|
||||
}
|
||||
|
||||
currentMessage = null;
|
||||
}
|
||||
},
|
||||
|
||||
flush(controller) {
|
||||
if (buffer) {
|
||||
try {
|
||||
const message = JSON.parse(buffer);
|
||||
controller.enqueue(message.content || '');
|
||||
} catch (e) {
|
||||
console.error('Failed to parse remaining buffer:', e);
|
||||
}
|
||||
}
|
||||
}
|
||||
});
|
||||
|
||||
// Connect the input to our transform stream
|
||||
return input.pipeThrough(transform);
|
||||
}
|
||||
|
||||
// Implementing the synchronous chat interaction
|
||||
public async chat(optionsArg: ChatOptions): Promise<ChatResponse> {
|
||||
// Format messages for Ollama
|
||||
const messages = [
|
||||
{ role: 'system', content: optionsArg.systemMessage },
|
||||
...optionsArg.messageHistory,
|
||||
{ role: 'user', content: optionsArg.userMessage }
|
||||
];
|
||||
|
||||
// Make API call to Ollama
|
||||
const response = await fetch(`${this.baseUrl}/api/chat`, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
body: JSON.stringify({
|
||||
model: this.model,
|
||||
messages: messages,
|
||||
stream: false
|
||||
}),
|
||||
});
|
||||
|
||||
if (!response.ok) {
|
||||
throw new Error(`Ollama API error: ${response.statusText}`);
|
||||
}
|
||||
|
||||
const result = await response.json();
|
||||
|
||||
return {
|
||||
role: 'assistant' as const,
|
||||
message: result.message.content,
|
||||
};
|
||||
}
|
||||
|
||||
public async audio(optionsArg: { message: string }): Promise<NodeJS.ReadableStream> {
|
||||
throw new Error('Audio generation is not supported by Ollama.');
|
||||
}
|
||||
|
||||
public async vision(optionsArg: { image: Buffer; prompt: string }): Promise<string> {
|
||||
const base64Image = optionsArg.image.toString('base64');
|
||||
|
||||
const response = await fetch(`${this.baseUrl}/api/chat`, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
body: JSON.stringify({
|
||||
model: this.visionModel,
|
||||
messages: [{
|
||||
role: 'user',
|
||||
content: optionsArg.prompt,
|
||||
images: [base64Image]
|
||||
}],
|
||||
stream: false
|
||||
}),
|
||||
});
|
||||
|
||||
if (!response.ok) {
|
||||
throw new Error(`Ollama API error: ${response.statusText}`);
|
||||
}
|
||||
|
||||
const result = await response.json();
|
||||
return result.message.content;
|
||||
}
|
||||
|
||||
public async document(optionsArg: {
|
||||
systemMessage: string;
|
||||
userMessage: string;
|
||||
pdfDocuments: Uint8Array[];
|
||||
messageHistory: ChatMessage[];
|
||||
}): Promise<{ message: any }> {
|
||||
// Convert PDF documents to images using SmartPDF
|
||||
const smartpdfInstance = new plugins.smartpdf.SmartPdf();
|
||||
let documentImageBytesArray: Uint8Array[] = [];
|
||||
|
||||
for (const pdfDocument of optionsArg.pdfDocuments) {
|
||||
const documentImageArray = await smartpdfInstance.convertPDFToPngBytes(pdfDocument);
|
||||
documentImageBytesArray = documentImageBytesArray.concat(documentImageArray);
|
||||
}
|
||||
|
||||
// Convert images to base64
|
||||
const base64Images = documentImageBytesArray.map(bytes => Buffer.from(bytes).toString('base64'));
|
||||
|
||||
// Send request to Ollama with images
|
||||
const response = await fetch(`${this.baseUrl}/api/chat`, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
body: JSON.stringify({
|
||||
model: this.visionModel,
|
||||
messages: [
|
||||
{ role: 'system', content: optionsArg.systemMessage },
|
||||
...optionsArg.messageHistory,
|
||||
{
|
||||
role: 'user',
|
||||
content: optionsArg.userMessage,
|
||||
images: base64Images
|
||||
}
|
||||
],
|
||||
stream: false
|
||||
}),
|
||||
});
|
||||
|
||||
if (!response.ok) {
|
||||
throw new Error(`Ollama API error: ${response.statusText}`);
|
||||
}
|
||||
|
||||
const result = await response.json();
|
||||
return {
|
||||
message: {
|
||||
role: 'assistant',
|
||||
content: result.message.content
|
||||
}
|
||||
};
|
||||
}
|
||||
}
|
||||
@ -1,10 +1,20 @@
|
||||
import * as plugins from './plugins.js';
|
||||
import * as paths from './paths.js';
|
||||
|
||||
// Custom type definition for chat completion messages
|
||||
export type TChatCompletionRequestMessage = {
|
||||
role: "system" | "user" | "assistant";
|
||||
content: string;
|
||||
};
|
||||
|
||||
import { MultiModalModel } from './abstract.classes.multimodal.js';
|
||||
|
||||
export interface IOpenaiProviderOptions {
|
||||
openaiToken: string;
|
||||
chatModel?: string;
|
||||
audioModel?: string;
|
||||
visionModel?: string;
|
||||
// Optionally add more model options (e.g., documentModel) if needed.
|
||||
}
|
||||
|
||||
export class OpenAiProvider extends MultiModalModel {
|
||||
@ -27,11 +37,79 @@ export class OpenAiProvider extends MultiModalModel {
|
||||
|
||||
public async stop() {}
|
||||
|
||||
public async chatStream(input: ReadableStream<string>): Promise<ReadableStream<string>> {
|
||||
// TODO: implement for OpenAI
|
||||
public async chatStream(input: ReadableStream<Uint8Array>): Promise<ReadableStream<string>> {
|
||||
// Create a TextDecoder to handle incoming chunks
|
||||
const decoder = new TextDecoder();
|
||||
let buffer = '';
|
||||
let currentMessage: {
|
||||
role: "function" | "user" | "system" | "assistant" | "tool" | "developer";
|
||||
content: string;
|
||||
} | null = null;
|
||||
|
||||
const returnStream = new ReadableStream();
|
||||
return returnStream;
|
||||
// Create a TransformStream to process the input
|
||||
const transform = new TransformStream<Uint8Array, string>({
|
||||
transform: async (chunk, controller) => {
|
||||
buffer += decoder.decode(chunk, { stream: true });
|
||||
|
||||
// Try to parse complete JSON messages from the buffer
|
||||
while (true) {
|
||||
const newlineIndex = buffer.indexOf('\n');
|
||||
if (newlineIndex === -1) break;
|
||||
|
||||
const line = buffer.slice(0, newlineIndex);
|
||||
buffer = buffer.slice(newlineIndex + 1);
|
||||
|
||||
if (line.trim()) {
|
||||
try {
|
||||
const message = JSON.parse(line);
|
||||
currentMessage = {
|
||||
role: (message.role || 'user') as "function" | "user" | "system" | "assistant" | "tool" | "developer",
|
||||
content: message.content || '',
|
||||
};
|
||||
} catch (e) {
|
||||
console.error('Failed to parse message:', e);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// If we have a complete message, send it to OpenAI
|
||||
if (currentMessage) {
|
||||
const messageToSend = { role: "user" as const, content: currentMessage.content };
|
||||
const chatModel = this.options.chatModel ?? 'o3-mini';
|
||||
const requestParams: any = {
|
||||
model: chatModel,
|
||||
messages: [messageToSend],
|
||||
stream: true,
|
||||
};
|
||||
// Temperature is omitted since the model does not support it.
|
||||
const stream = await this.openAiApiClient.chat.completions.create(requestParams);
|
||||
// Explicitly cast the stream as an async iterable to satisfy TypeScript.
|
||||
const streamAsyncIterable = stream as unknown as AsyncIterableIterator<any>;
|
||||
// Process each chunk from OpenAI
|
||||
for await (const chunk of streamAsyncIterable) {
|
||||
const content = chunk.choices[0]?.delta?.content;
|
||||
if (content) {
|
||||
controller.enqueue(content);
|
||||
}
|
||||
}
|
||||
currentMessage = null;
|
||||
}
|
||||
},
|
||||
|
||||
flush(controller) {
|
||||
if (buffer) {
|
||||
try {
|
||||
const message = JSON.parse(buffer);
|
||||
controller.enqueue(message.content || '');
|
||||
} catch (e) {
|
||||
console.error('Failed to parse remaining buffer:', e);
|
||||
}
|
||||
}
|
||||
}
|
||||
});
|
||||
|
||||
// Connect the input to our transform stream
|
||||
return input.pipeThrough(transform);
|
||||
}
|
||||
|
||||
// Implementing the synchronous chat interaction
|
||||
@ -43,15 +121,17 @@ export class OpenAiProvider extends MultiModalModel {
|
||||
content: string;
|
||||
}[];
|
||||
}) {
|
||||
const result = await this.openAiApiClient.chat.completions.create({
|
||||
model: 'gpt-4-turbo-preview',
|
||||
|
||||
const chatModel = this.options.chatModel ?? 'o3-mini';
|
||||
const requestParams: any = {
|
||||
model: chatModel,
|
||||
messages: [
|
||||
{ role: 'system', content: optionsArg.systemMessage },
|
||||
...optionsArg.messageHistory,
|
||||
{ role: 'user', content: optionsArg.userMessage },
|
||||
],
|
||||
});
|
||||
};
|
||||
// Temperature parameter removed to avoid unsupported error.
|
||||
const result = await this.openAiApiClient.chat.completions.create(requestParams);
|
||||
return {
|
||||
role: result.choices[0].message.role as 'assistant',
|
||||
message: result.choices[0].message.content,
|
||||
@ -61,7 +141,7 @@ export class OpenAiProvider extends MultiModalModel {
|
||||
public async audio(optionsArg: { message: string }): Promise<NodeJS.ReadableStream> {
|
||||
const done = plugins.smartpromise.defer<NodeJS.ReadableStream>();
|
||||
const result = await this.openAiApiClient.audio.speech.create({
|
||||
model: 'tts-1-hd',
|
||||
model: this.options.audioModel ?? 'o3-mini',
|
||||
input: optionsArg.message,
|
||||
voice: 'nova',
|
||||
response_format: 'mp3',
|
||||
@ -83,27 +163,30 @@ export class OpenAiProvider extends MultiModalModel {
|
||||
}) {
|
||||
let pdfDocumentImageBytesArray: Uint8Array[] = [];
|
||||
|
||||
// Convert each PDF into one or more image byte arrays.
|
||||
const smartpdfInstance = new plugins.smartpdf.SmartPdf();
|
||||
await smartpdfInstance.start();
|
||||
for (const pdfDocument of optionsArg.pdfDocuments) {
|
||||
const documentImageArray = await this.smartpdfInstance.convertPDFToPngBytes(pdfDocument);
|
||||
const documentImageArray = await smartpdfInstance.convertPDFToPngBytes(pdfDocument);
|
||||
pdfDocumentImageBytesArray = pdfDocumentImageBytesArray.concat(documentImageArray);
|
||||
}
|
||||
await smartpdfInstance.stop();
|
||||
|
||||
console.log(`image smartfile array`);
|
||||
console.log(pdfDocumentImageBytesArray.map((smartfile) => smartfile.length));
|
||||
|
||||
const smartfileArray = await plugins.smartarray.map(
|
||||
pdfDocumentImageBytesArray,
|
||||
async (pdfDocumentImageBytes) => {
|
||||
return plugins.smartfile.SmartFile.fromBuffer(
|
||||
'pdfDocumentImage.jpg',
|
||||
Buffer.from(pdfDocumentImageBytes)
|
||||
);
|
||||
}
|
||||
);
|
||||
// Filter out any empty buffers to avoid sending invalid image URLs.
|
||||
const validImageBytesArray = pdfDocumentImageBytesArray.filter(imageBytes => imageBytes && imageBytes.length > 0);
|
||||
const imageAttachments = validImageBytesArray.map(imageBytes => ({
|
||||
type: 'image_url',
|
||||
image_url: {
|
||||
url: 'data:image/png;base64,' + Buffer.from(imageBytes).toString('base64'),
|
||||
},
|
||||
}));
|
||||
|
||||
const result = await this.openAiApiClient.chat.completions.create({
|
||||
model: 'gpt-4-vision-preview',
|
||||
// response_format: { type: "json_object" }, // not supported for now
|
||||
const chatModel = this.options.chatModel ?? 'gpt-4o';
|
||||
const requestParams: any = {
|
||||
model: chatModel,
|
||||
messages: [
|
||||
{ role: 'system', content: optionsArg.systemMessage },
|
||||
...optionsArg.messageHistory,
|
||||
@ -111,24 +194,39 @@ export class OpenAiProvider extends MultiModalModel {
|
||||
role: 'user',
|
||||
content: [
|
||||
{ type: 'text', text: optionsArg.userMessage },
|
||||
...(() => {
|
||||
const returnArray = [];
|
||||
for (const imageBytes of pdfDocumentImageBytesArray) {
|
||||
returnArray.push({
|
||||
type: 'image_url',
|
||||
image_url: {
|
||||
url: 'data:image/png;base64,' + Buffer.from(imageBytes).toString('base64'),
|
||||
},
|
||||
});
|
||||
}
|
||||
return returnArray;
|
||||
})(),
|
||||
...imageAttachments,
|
||||
],
|
||||
},
|
||||
],
|
||||
});
|
||||
};
|
||||
// Temperature parameter removed.
|
||||
const result = await this.openAiApiClient.chat.completions.create(requestParams);
|
||||
return {
|
||||
message: result.choices[0].message,
|
||||
};
|
||||
}
|
||||
|
||||
public async vision(optionsArg: { image: Buffer; prompt: string }): Promise<string> {
|
||||
const visionModel = this.options.visionModel ?? 'gpt-4o';
|
||||
const requestParams: any = {
|
||||
model: visionModel,
|
||||
messages: [
|
||||
{
|
||||
role: 'user',
|
||||
content: [
|
||||
{ type: 'text', text: optionsArg.prompt },
|
||||
{
|
||||
type: 'image_url',
|
||||
image_url: {
|
||||
url: `data:image/jpeg;base64,${optionsArg.image.toString('base64')}`
|
||||
}
|
||||
}
|
||||
]
|
||||
}
|
||||
],
|
||||
max_tokens: 300
|
||||
};
|
||||
const result = await this.openAiApiClient.chat.completions.create(requestParams);
|
||||
return result.choices[0].message.content || '';
|
||||
}
|
||||
}
|
||||
@ -1,3 +1,171 @@
|
||||
import * as plugins from './plugins.js';
|
||||
import * as paths from './paths.js';
|
||||
import { MultiModalModel } from './abstract.classes.multimodal.js';
|
||||
import type { ChatOptions, ChatResponse, ChatMessage } from './abstract.classes.multimodal.js';
|
||||
|
||||
export class PerplexityProvider {}
|
||||
export interface IPerplexityProviderOptions {
|
||||
perplexityToken: string;
|
||||
}
|
||||
|
||||
export class PerplexityProvider extends MultiModalModel {
|
||||
private options: IPerplexityProviderOptions;
|
||||
|
||||
constructor(optionsArg: IPerplexityProviderOptions) {
|
||||
super();
|
||||
this.options = optionsArg;
|
||||
}
|
||||
|
||||
async start() {
|
||||
// Initialize any necessary clients or resources
|
||||
}
|
||||
|
||||
async stop() {}
|
||||
|
||||
public async chatStream(input: ReadableStream<Uint8Array>): Promise<ReadableStream<string>> {
|
||||
// Create a TextDecoder to handle incoming chunks
|
||||
const decoder = new TextDecoder();
|
||||
let buffer = '';
|
||||
let currentMessage: { role: string; content: string; } | null = null;
|
||||
|
||||
// Create a TransformStream to process the input
|
||||
const transform = new TransformStream<Uint8Array, string>({
|
||||
async transform(chunk, controller) {
|
||||
buffer += decoder.decode(chunk, { stream: true });
|
||||
|
||||
// Try to parse complete JSON messages from the buffer
|
||||
while (true) {
|
||||
const newlineIndex = buffer.indexOf('\n');
|
||||
if (newlineIndex === -1) break;
|
||||
|
||||
const line = buffer.slice(0, newlineIndex);
|
||||
buffer = buffer.slice(newlineIndex + 1);
|
||||
|
||||
if (line.trim()) {
|
||||
try {
|
||||
const message = JSON.parse(line);
|
||||
currentMessage = {
|
||||
role: message.role || 'user',
|
||||
content: message.content || '',
|
||||
};
|
||||
} catch (e) {
|
||||
console.error('Failed to parse message:', e);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// If we have a complete message, send it to Perplexity
|
||||
if (currentMessage) {
|
||||
const response = await fetch('https://api.perplexity.ai/chat/completions', {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Authorization': `Bearer ${this.options.perplexityToken}`,
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
body: JSON.stringify({
|
||||
model: 'mixtral-8x7b-instruct',
|
||||
messages: [{ role: currentMessage.role, content: currentMessage.content }],
|
||||
stream: true,
|
||||
}),
|
||||
});
|
||||
|
||||
// Process each chunk from Perplexity
|
||||
const reader = response.body?.getReader();
|
||||
if (reader) {
|
||||
try {
|
||||
while (true) {
|
||||
const { done, value } = await reader.read();
|
||||
if (done) break;
|
||||
|
||||
const chunk = new TextDecoder().decode(value);
|
||||
const lines = chunk.split('\n');
|
||||
|
||||
for (const line of lines) {
|
||||
if (line.startsWith('data: ')) {
|
||||
const data = line.slice(6);
|
||||
if (data === '[DONE]') break;
|
||||
|
||||
try {
|
||||
const parsed = JSON.parse(data);
|
||||
const content = parsed.choices[0]?.delta?.content;
|
||||
if (content) {
|
||||
controller.enqueue(content);
|
||||
}
|
||||
} catch (e) {
|
||||
console.error('Failed to parse SSE data:', e);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
} finally {
|
||||
reader.releaseLock();
|
||||
}
|
||||
}
|
||||
|
||||
currentMessage = null;
|
||||
}
|
||||
},
|
||||
|
||||
flush(controller) {
|
||||
if (buffer) {
|
||||
try {
|
||||
const message = JSON.parse(buffer);
|
||||
controller.enqueue(message.content || '');
|
||||
} catch (e) {
|
||||
console.error('Failed to parse remaining buffer:', e);
|
||||
}
|
||||
}
|
||||
}
|
||||
});
|
||||
|
||||
// Connect the input to our transform stream
|
||||
return input.pipeThrough(transform);
|
||||
}
|
||||
|
||||
// Implementing the synchronous chat interaction
|
||||
public async chat(optionsArg: ChatOptions): Promise<ChatResponse> {
|
||||
// Make API call to Perplexity
|
||||
const response = await fetch('https://api.perplexity.ai/chat/completions', {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Authorization': `Bearer ${this.options.perplexityToken}`,
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
body: JSON.stringify({
|
||||
model: 'mixtral-8x7b-instruct', // Using Mixtral model
|
||||
messages: [
|
||||
{ role: 'system', content: optionsArg.systemMessage },
|
||||
...optionsArg.messageHistory,
|
||||
{ role: 'user', content: optionsArg.userMessage }
|
||||
],
|
||||
}),
|
||||
});
|
||||
|
||||
if (!response.ok) {
|
||||
throw new Error(`Perplexity API error: ${response.statusText}`);
|
||||
}
|
||||
|
||||
const result = await response.json();
|
||||
|
||||
return {
|
||||
role: 'assistant' as const,
|
||||
message: result.choices[0].message.content,
|
||||
};
|
||||
}
|
||||
|
||||
public async audio(optionsArg: { message: string }): Promise<NodeJS.ReadableStream> {
|
||||
throw new Error('Audio generation is not supported by Perplexity.');
|
||||
}
|
||||
|
||||
public async vision(optionsArg: { image: Buffer; prompt: string }): Promise<string> {
|
||||
throw new Error('Vision tasks are not supported by Perplexity.');
|
||||
}
|
||||
|
||||
public async document(optionsArg: {
|
||||
systemMessage: string;
|
||||
userMessage: string;
|
||||
pdfDocuments: Uint8Array[];
|
||||
messageHistory: ChatMessage[];
|
||||
}): Promise<{ message: any }> {
|
||||
throw new Error('Document processing is not supported by Perplexity.');
|
||||
}
|
||||
}
|
||||
183
ts/provider.xai.ts
Normal file
183
ts/provider.xai.ts
Normal file
@ -0,0 +1,183 @@
|
||||
import * as plugins from './plugins.js';
|
||||
import * as paths from './paths.js';
|
||||
import { MultiModalModel } from './abstract.classes.multimodal.js';
|
||||
import type { ChatOptions, ChatResponse, ChatMessage } from './abstract.classes.multimodal.js';
|
||||
import type { ChatCompletionMessageParam } from 'openai/resources/chat/completions';
|
||||
|
||||
export interface IXAIProviderOptions {
|
||||
xaiToken: string;
|
||||
}
|
||||
|
||||
export class XAIProvider extends MultiModalModel {
|
||||
private options: IXAIProviderOptions;
|
||||
public openAiApiClient: plugins.openai.default;
|
||||
public smartpdfInstance: plugins.smartpdf.SmartPdf;
|
||||
|
||||
constructor(optionsArg: IXAIProviderOptions) {
|
||||
super();
|
||||
this.options = optionsArg;
|
||||
}
|
||||
|
||||
public async start() {
|
||||
this.openAiApiClient = new plugins.openai.default({
|
||||
apiKey: this.options.xaiToken,
|
||||
baseURL: 'https://api.x.ai/v1',
|
||||
});
|
||||
this.smartpdfInstance = new plugins.smartpdf.SmartPdf();
|
||||
}
|
||||
|
||||
public async stop() {}
|
||||
|
||||
public async chatStream(input: ReadableStream<Uint8Array>): Promise<ReadableStream<string>> {
|
||||
// Create a TextDecoder to handle incoming chunks
|
||||
const decoder = new TextDecoder();
|
||||
let buffer = '';
|
||||
let currentMessage: { role: string; content: string; } | null = null;
|
||||
|
||||
// Create a TransformStream to process the input
|
||||
const transform = new TransformStream<Uint8Array, string>({
|
||||
async transform(chunk, controller) {
|
||||
buffer += decoder.decode(chunk, { stream: true });
|
||||
|
||||
// Try to parse complete JSON messages from the buffer
|
||||
while (true) {
|
||||
const newlineIndex = buffer.indexOf('\n');
|
||||
if (newlineIndex === -1) break;
|
||||
|
||||
const line = buffer.slice(0, newlineIndex);
|
||||
buffer = buffer.slice(newlineIndex + 1);
|
||||
|
||||
if (line.trim()) {
|
||||
try {
|
||||
const message = JSON.parse(line);
|
||||
currentMessage = {
|
||||
role: message.role || 'user',
|
||||
content: message.content || '',
|
||||
};
|
||||
} catch (e) {
|
||||
console.error('Failed to parse message:', e);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// If we have a complete message, send it to X.AI
|
||||
if (currentMessage) {
|
||||
const stream = await this.openAiApiClient.chat.completions.create({
|
||||
model: 'grok-2-latest',
|
||||
messages: [{ role: currentMessage.role, content: currentMessage.content }],
|
||||
stream: true,
|
||||
});
|
||||
|
||||
// Process each chunk from X.AI
|
||||
for await (const chunk of stream) {
|
||||
const content = chunk.choices[0]?.delta?.content;
|
||||
if (content) {
|
||||
controller.enqueue(content);
|
||||
}
|
||||
}
|
||||
|
||||
currentMessage = null;
|
||||
}
|
||||
},
|
||||
|
||||
flush(controller) {
|
||||
if (buffer) {
|
||||
try {
|
||||
const message = JSON.parse(buffer);
|
||||
controller.enqueue(message.content || '');
|
||||
} catch (e) {
|
||||
console.error('Failed to parse remaining buffer:', e);
|
||||
}
|
||||
}
|
||||
}
|
||||
});
|
||||
|
||||
// Connect the input to our transform stream
|
||||
return input.pipeThrough(transform);
|
||||
}
|
||||
|
||||
public async chat(optionsArg: {
|
||||
systemMessage: string;
|
||||
userMessage: string;
|
||||
messageHistory: { role: string; content: string; }[];
|
||||
}): Promise<{ role: 'assistant'; message: string; }> {
|
||||
// Prepare messages array with system message, history, and user message
|
||||
const messages: ChatCompletionMessageParam[] = [
|
||||
{ role: 'system', content: optionsArg.systemMessage },
|
||||
...optionsArg.messageHistory.map(msg => ({
|
||||
role: msg.role as 'system' | 'user' | 'assistant',
|
||||
content: msg.content
|
||||
})),
|
||||
{ role: 'user', content: optionsArg.userMessage }
|
||||
];
|
||||
|
||||
// Call X.AI's chat completion API
|
||||
const completion = await this.openAiApiClient.chat.completions.create({
|
||||
model: 'grok-2-latest',
|
||||
messages: messages,
|
||||
stream: false,
|
||||
});
|
||||
|
||||
// Return the assistant's response
|
||||
return {
|
||||
role: 'assistant',
|
||||
message: completion.choices[0]?.message?.content || ''
|
||||
};
|
||||
}
|
||||
|
||||
public async audio(optionsArg: { message: string }): Promise<NodeJS.ReadableStream> {
|
||||
throw new Error('Audio generation is not supported by X.AI');
|
||||
}
|
||||
|
||||
public async vision(optionsArg: { image: Buffer; prompt: string }): Promise<string> {
|
||||
throw new Error('Vision tasks are not supported by X.AI');
|
||||
}
|
||||
|
||||
public async document(optionsArg: {
|
||||
systemMessage: string;
|
||||
userMessage: string;
|
||||
pdfDocuments: Uint8Array[];
|
||||
messageHistory: { role: string; content: string; }[];
|
||||
}): Promise<{ message: any }> {
|
||||
// First convert PDF documents to images
|
||||
let pdfDocumentImageBytesArray: Uint8Array[] = [];
|
||||
|
||||
for (const pdfDocument of optionsArg.pdfDocuments) {
|
||||
const documentImageArray = await this.smartpdfInstance.convertPDFToPngBytes(pdfDocument);
|
||||
pdfDocumentImageBytesArray = pdfDocumentImageBytesArray.concat(documentImageArray);
|
||||
}
|
||||
|
||||
// Convert images to base64 for inclusion in the message
|
||||
const imageBase64Array = pdfDocumentImageBytesArray.map(bytes =>
|
||||
Buffer.from(bytes).toString('base64')
|
||||
);
|
||||
|
||||
// Combine document images into the user message
|
||||
const enhancedUserMessage = `
|
||||
${optionsArg.userMessage}
|
||||
|
||||
Document contents (as images):
|
||||
${imageBase64Array.map((img, i) => `Image ${i + 1}: <image data>`).join('\n')}
|
||||
`;
|
||||
|
||||
// Use chat completion to analyze the documents
|
||||
const messages: ChatCompletionMessageParam[] = [
|
||||
{ role: 'system', content: optionsArg.systemMessage },
|
||||
...optionsArg.messageHistory.map(msg => ({
|
||||
role: msg.role as 'system' | 'user' | 'assistant',
|
||||
content: msg.content
|
||||
})),
|
||||
{ role: 'user', content: enhancedUserMessage }
|
||||
];
|
||||
|
||||
const completion = await this.openAiApiClient.chat.completions.create({
|
||||
model: 'grok-2-latest',
|
||||
messages: messages,
|
||||
stream: false,
|
||||
});
|
||||
|
||||
return {
|
||||
message: completion.choices[0]?.message?.content || ''
|
||||
};
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user