prism_ai 0.6.0

Lean, provider-agnostic streaming client for AI chat completions — OpenAI, Ollama, Groq, DeepSeek and more, with typed events.

Prism AI

A lean, provider-agnostic streaming client for AI chat completions. One package covers OpenAI, Ollama, Groq, DeepSeek, Together, LM Studio, vLLM and every other OpenAI-compatible endpoint — with typed streaming events and zero heavyweight dependencies.

Part of the Prism family.

Usage

final client = OpenAiStreamClient(
  baseUrl: 'http://localhost:11434/v1', // Ollama
  // baseUrl: 'https://api.openai.com/v1', apiKey: 'sk-...',
);

await for (final event in client.streamChat(messages, model: 'llama3')) {
  switch (event) {
    case AiTextDelta(:final text):
      setState(() => response += text);   // stream into PrismMarkdown!
    case AiToolCall(:final name):
      handleTool(name);
    case AiDone(:final usage):
      log('done: ${usage?.outputTokens} tokens');
    case AiError(:final message):
      showError(message);
  }
}

Why prism_ai

	prism_ai	dart_openai	langchain_dart
Streaming SSE with typed events	✅	❌	✅
One client, any OpenAI-compatible host	✅	❌	partial
Lightweight (http only)	✅	❌	❌ heavy
Maintained	✅	❌ abandoned	✅

Pairs with prism_markdown: render streaming AI responses as live markdown.

Architecture

• SseParser — stateful incremental SSE parser (chunk-safe, CRLF-safe)
• OpenAiStreamClient — streams /chat/completions with auth + headers
• Typed events: AiTextDelta, AiToolCall, AiDone, AiError

Providers

Provider	Client	Notes
OpenAI & every OpenAI-compatible host	OpenAiStreamClient	OpenAI, Ollama, Groq, DeepSeek, Together, LM Studio, vLLM, Azure
Anthropic (Claude)	AnthropicStreamClient	Messages API with tool-use streaming
Google Gemini	GeminiStreamClient	streamGenerateContent (SSE)

Chat UI included

AiChatView renders a full streaming chat conversation from an AiChatController — user/assistant bubbles, typing indicator, auto-scroll:

final controller = AiChatController();
AiChatView(
  controller: controller,
  onSendMessage: (text) async {
    controller.sendUserMessage(text);
    await for (final event in client.streamChat(
      controller.messages, model: 'llama3')) {
      switch (event) {
        case AiTextDelta(:final text):
          controller.appendAssistantDelta(text);
        case AiDone():
          controller.finishAssistant();
        case AiError(:final message):
          controller.reportError(message);
      }
    }
  },
)

Roadmap

• ❌ Embeddings API support
• ❌ Markdown rendering inside assistant bubbles via prism_markdown

License

MIT