@agentsy/core/processor

• Status: Published subpath export from @agentsy/core
• Role: Core stream processing primitives and provider-agnostic transforms

Where it fits

@agentsy/core/processor is the center of the current streaming stack. It consumes normalized events and produces a stable flow that downstream renderers, stores, and agent loops can use.

Key exports

• LLMStreamProcessor
• createProcessorEventAdapter
• ToolCallParser
• ZAiInlineToolCallParser
• createSmoothStream
• createThinkingFilter
• createToolCallFilter

Available APIs

• Processor runtime: LLMStreamProcessor, createProcessorEventAdapter
• Tool-call parsing helpers: ToolCallParser, ZAiInlineToolCallParser
• Processor transforms: createSmoothStream, createThinkingFilter, createToolCallFilter
• SSE helpers re-exported from the processor package surface

Use it when

• you already have provider chunks normalized and need an incremental processor
• you want reusable stream transforms instead of bespoke event glue
• you need the main orchestration layer beneath UI or agent code

Common neighbors

• Upstream: @agentsy/providers/normalizers, @agentsy/providers/pipeline
• Side utilities: @agentsy/core/thinking, @agentsy/core/tool-calls, @agentsy/core/structured
• Downstream: @agentsy/renderers, @agentsy/ui, @agentsy/orchestrator/agent, @agentsy/vscode

Example

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

const processor = new LLMStreamProcessor();
for await (const event of createPipeline(openAiSseStream, {
  provider: 'openai'
})) {
  console.log(event);
}

Implementation example with neighbors

import { normalizeAnthropicEvent } from '@agentsy/providers/normalizers';
import { LLMStreamProcessor } from '@agentsy/core/processor';
import { createConversationStoreFromProcessor } from '@agentsy/ui';

const processor = new LLMStreamProcessor({ parseThinkTags: true });
const bridge = createConversationStoreFromProcessor(processor, {
  conversationId: 'demo-conversation'
});

for await (const event of anthropicEventStream) {
  processor.process(normalizeAnthropicEvent(event));
}

console.log(bridge.store.getState().messages.length);
bridge.dispose();

Notes

Use @agentsy/core/processor for the processor itself, and pair it with @agentsy/providers/* when you need provider-specific normalizers, adapters, or SSE pipelines.