@agentsy/providers/normalizers

• Status: Published subpath export from @agentsy/providers
• Role: Provider-specific normalization into a shared stream vocabulary

Where it fits

@agentsy/providers/normalizers is the compatibility layer between raw provider payloads and the rest of the Agentsy processing stack.

Key exports

• normalizeOpenAIChatChunk
• normalizeOpenAIResponseEvent
• normalizeAnthropicEvent
• normalizeGeminiChunk
• normalizeMistralChunk
• normalizeCohereEvent
• normalizeOllamaChatChunk
• normalizeOllamaGenerateChunk
• normalizeBedrockConverseEvent
• normalizeHuggingFaceTGIChunk
• normalizeDeepSeekChunk
• normalizeZAiChunk
• OpenAI-compatible helpers

Available APIs

• Provider-specific normalizers for Anthropic, Bedrock, Cohere, DeepSeek, Gemini, Hugging Face TGI, Mistral, Ollama, OpenAI, and Z.AI
• OpenAI-compatible provider detection and normalization helpers
• Shared normalizer types re-exported from the package

Use it when

• you need one downstream pipeline to support multiple providers
• you want the rest of your code to stop caring about provider wire formats

Common neighbors

• Downstream: @agentsy/core/processor, @agentsy/providers/pipeline
• Adjacent docs: Stream processing flow

Example

import { normalizeOpenAIChatChunk } from '@agentsy/providers/normalizers';

const normalized = normalizeOpenAIChatChunk(chunk);

Implementation example with neighbors

import { normalizeOpenAIChatChunk } from '@agentsy/providers/normalizers';
import { LLMStreamProcessor } from '@agentsy/core/processor';

const processor = new LLMStreamProcessor({ parseThinkTags: true });

for await (const rawChunk of openAiStream) {
  processor.process(normalizeOpenAIChatChunk(rawChunk));
}

console.log(processor.accumulatedMessage.content);

Notes

If your integration already emits a compatible internal event shape, you may not need this package directly. Most provider-facing code does.