NewHow the Agent Session Protocol builds trust between autonomous systems.
Product

Raw data comes in. Clean, enriched, classified data goes out.

Transform, enrich, classify, route, and deliver. Build the entire pipeline on one visual canvas. No scripts. No cron jobs. No glue code.

A webhook fires with a raw order payload.

You need to enrich it with customer data from your CRM.

Classify it by risk level using your LLM.

Transform the schema for your data warehouse.

Route high risk orders to fraud review.

Deliver the rest to fulfillment in real time.

Today, that takes five microservices and a prayer.

One fails silently. Nobody knows until a customer complains.

Pipes puts all of it on one visual canvas.

Every step visible. Every failure recovered. Every event accounted for.

Enrich. Transform. Route. Deliver. One canvas.

Raw data arrives from a webhook. Enriched with CRM data. Classified by your LLM. Transformed for your warehouse. Routed by business rules. All on one visual canvas.

Classify, extract, and route with your own LLM.

Drop an AI node at any step. It sends the event to your LLM with your prompt, and outputs structured data to the next node. Risk classification, entity extraction, intelligent routing.

Classify by risk

Every order is scored by your LLM. High value, high risk, suspicious. Risky orders are routed to fraud review automatically.

Extract what matters

Names, addresses, amounts, references. The AI extracts structured data from free text, PDFs, and webhook bodies.

Route intelligently

Approved orders go to fulfillment. Suspicious orders go to review. Transformed data goes to the warehouse. All automatic.

Multi step processes that recover automatically.

When a step fails, compensation actions run in reverse for all completed steps. No manual intervention.

Saga with automatic compensation
Step 1
Reserve inventory
completed
Step 2
Charge payment
completed
Step 3
Book shipping
failed
Compensating
Refund payment
reversed
Release inventory
reversed
System restored. Zero manual intervention.

Resume from exactly where it stopped.

Every node saves state. No reprocessing. No skipping.

1

Checkpoint

Workflow state is saved at every node.

2

Failure

The next node fails. The pipeline stops.

3

Resume

The system resumes from the exact checkpoint.

4

Complete

The pipeline finishes normally. Zero data lost.

Before and after Pipes

Without Pipes

  • Five microservices for one pipeline
  • Each one can fail independently
  • No visual overview of data flow
  • No single place to debug
  • Manual error recovery
  • Separate monitoring per service

With Pipes

  • One canvas with every step visible
  • Built in error handling per node
  • Real time visual overview
  • Unified debugging in the dashboard
  • Saga compensation runs automatically
  • Per node observability

npayload Pipes vs. building pipelines yourself

FeaturenpayloadBuild it yourself
Visual builder with versioning
AI classification nodesMonths of work
Saga with auto compensationComplex to build
Checkpointing and resumeWeeks of work
Per node observabilitySeparate project
Pre built connectorsBuild each one
DLQ with replayWeeks of work
Automatic backpressureComplex to build

Frequently asked questions

Do I need to write code to use Pipes?+
No. Visual nodes handle most workflows. Add custom code nodes when you need advanced logic.
How do AI nodes work in practice?+
An AI node sends the event to your LLM with your prompt and outputs structured results to the next node.
What happens when a saga step fails?+
Compensation actions run in reverse for all completed steps automatically. No manual intervention.
Can I monitor pipelines in real time?+
Yes. Per node latency, throughput, error rates, and full event journey tracing in the dashboard.
How does this compare to Airflow or Step Functions?+
Pipes processes events in real time, not on a schedule. Sagas provide distributed transactions that neither offers natively.
What about backpressure when a slow node bottlenecks the pipeline?+
Automatic buffering and flow adjustment. Events are never dropped. Alerts fire if backpressure persists.
Can I version and rollback workflows?+
Yes. Every change is versioned. Rollback with one click. In flight events complete under the old version.