@agentsy/core/structured

• Status: Published subpath export from @agentsy/core
• Role: JSON parsing, repair, validation, and schema-oriented streaming helpers

Key exports

• parseJson
• validateJsonSchema
• buildFormatInstructions
• buildRepairPrompt
• streamJson
• autoRepair
• providerFormats
• repairStateMachine
• fieldValidator
• zodAdapter

Available APIs

• JSON parsing and schema validation
• Repair prompt and format-instruction builders
• Streaming JSON helpers and provider-format adapters

Where it fits

This package handles the part where model output almost looks like structured data but is still trying very hard to disappoint you.

Common neighbors

• @agentsy/core/processor
• @agentsy/core/tool-calls
• @agentsy/orchestrator/agent

Implementation example with neighbors

import { normalizeOpenAIChatChunk } from '@agentsy/providers/normalizers';
import { LLMStreamProcessor } from '@agentsy/core/processor';
import { parseJson, validateJsonSchema } from '@agentsy/core/structured';

const processor = new LLMStreamProcessor();

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

const content = processor.accumulatedMessage.content;
const payload = parseJson(content);
const validated = validateJsonSchema(content, responseSchema);
console.log(validated.success);