Platform connectors

How agent data gets into Muster

There are three integration paths. Most customers use OTel and platform connectors for full coverage from day one, then add the SDK to their most critical agents.

	OTel	Platform connectors	SDK
Best for	Custom-coded agents	No-code/low-code tools	Critical agents needing precise checks
Setup	3 env vars	Elitery deploys	Developer adds `@agent.govern()`
Quality signals	✅ Inferred by Muster	✅ Inferred by Muster	✅ Exact — you define the rules
Token / cost tracking	✅	✅	✅
Discovery	✅	✅	❌

Layer 1 — OTel (primary for coded agents)

For any agent you can configure, OTel is the recommended path. Set three env vars and Muster receives traces, infers quality checks, and tracks costs automatically.

OTEL_EXPORTER_OTLP_ENDPOINT=https://backend.YOUR-INSTANCE.getmuster.io/api/v1/otel
OTEL_EXPORTER_OTLP_HEADERS=Authorization=Bearer <muster-jwt-token>
OTEL_EXPORTER_OTLP_PROTOCOL=http/json

Muster’s check inference engine runs on every execution — it infers relevant checks from the agent name, output structure, historical patterns, and Claude Haiku fallback. No check code required. → Full OTel setup guide

Layer 2 — Platform connectors (for no-code tools)

For platforms where you can’t set env vars — visual workflow tools and managed AI services. Elitery deploys and manages all connectors. Same check inference engine applies.

Connector	What it covers	How it discovers
n8n	n8n workflows with LLM nodes	Workflow API scan
AWS Bedrock	Bedrock agents	`list_agents()` API
LangChain / LangSmith	LangChain agents	LangSmith project scan
Langfuse	Any agent traced to Langfuse	Trace scan
Azure AI	Azure OpenAI + AI Foundry	ARM API scan

Layer 3 — SDK (for critical agents)

The SDK adds precise, business-rule checks on top of OTel/connector coverage. Use it for agents where inferred checks aren’t enough — financial calculations, compliance decisions, external writes.

@agent.govern(
    required_fields=["vendor_name", "invoice_number", "total"],
    arithmetic=[{"check_id": "total_check", "result_field": "total",
                 "component_fields": ["subtotal", "tax_amount"], "operation": "sum"}],
)
async def process_invoice(input_data: dict) -> dict:
    return extracted_output  # your existing code, unchanged

→ SDK integration guide

Check inference — how inferred signals work

When OTel or a connector submits an execution, Muster infers quality checks automatically in four layers:

Name matching — agent name → check category (financial, legal, risk, support)
Schema analysis — output field names → check types (numeric, decision, required fields)
Historical patterns — after 50+ executions, distribution-based checks
LLM fallback — Claude generates check suggestions from agent name + output structure

Inferred checks are good enough for monitoring and trend detection. For agents where you need exact validation with specific expected values — add the SDK.

Recommended approach by agent type

Agent	Recommended path
Custom Python/Go/Node agent	OTel
LangChain, CrewAI, LlamaIndex	OTel (native support)
n8n, Flowise, Dify workflow	Platform connector
AWS Bedrock, Azure AI Foundry	Platform connector
Invoice / AP processing agent	OTel or connector + SDK
Compliance / legal review agent	OTel or connector + SDK
Fraud / risk screening agent	OTel or connector + SDK

​How agent data gets into Muster

​Layer 1 — OTel (primary for coded agents)

​Layer 2 — Platform connectors (for no-code tools)

​Layer 3 — SDK (for critical agents)

​Check inference — how inferred signals work

​Recommended approach by agent type