---
title: Managed Execution
description: How Edgaze operates retries, streaming, long-running state, usage records, and infrastructure behind every workflow run.
source: https://www.edgaze.ai/docs/managed-execution
section: resources
---
# Managed Execution

> How Edgaze operates retries, streaming, long-running state, usage records, and infrastructure behind every workflow run.

**Audience: Creators and buyers who want to understand what Edgaze operates during a run.**

**Outcome: Know what Edgaze handles for you so you can publish workflows without running your own execution stack.**

## Overview

When someone runs a published workflow, Edgaze operates the execution path end to end: scheduling the run, executing the graph, coordinating models and tools, applying retries where the platform supports them, streaming or polling progress, persisting run state and events, and writing usage and billing records.

You publish the product. Buyers and API callers trigger runs. Edgaze runs the infrastructure.

## What Edgaze operates

### Execution

Hosted runs execute on Edgaze runtime infrastructure. A run may include multiple nodes - inputs, models, tools, HTTP calls, conditions, merges, delays, and outputs - as defined in your published graph. See [Workflow Run Policy](/docs/workflow-run-policy).

### Retries and failover

Transient provider or tool failures may be retried according to platform behavior for that node and run. Creators can also control **model failover** at publish time: when failover is enabled, Edgaze may route to a fallback model if the primary is degraded; extra failover cost above the primary estimate is absorbed by Edgaze and does not change the buyer’s displayed price. When failover is disabled, a degraded primary fails the run without charging margin. See [Failover](/docs/billing/failover).

### Streaming and progress

Buyers see live progress in the product page run panel. Developers can poll run status, read ordered run events, or connect to the optional SSE stream for a run. Streaming and event APIs expose sanitized progress - not internal billing or raw provider payloads. See [Streaming & Real-time](/docs/api/streaming).

### Long-running state

Runs persist through the platform’s run lifecycle (queued, preparing, started, terminal states). Buyers and API clients can leave and return to a run’s status and results without keeping a browser tab open for the entire execution. Event sequences are durable for progress and debugging.

### Usage records and billing settlement

Every run produces records used for status, support, and settlement. Successful paid runs charge the displayed price (wallet or bundle). Failed or cancelled runs release margin charges; Edgaze absorbs ordinary compute on those outcomes. Demo and builder test runs are never billed. See [When Runs Fail](/docs/billing/when-runs-fail) and [Run Outcomes Reference](/docs/billing/run-outcomes-reference).

### Infrastructure you do not run

Creators do not provision GPUs, queue workers, or billing ledgers for marketplace or API runs. Publish the workflow; Edgaze hosts execution. Infrastructure cost estimates at publish time inform the compute portion of the displayed price - buyers still see one price, not a raw infra invoice. See [Infrastructure Cost Estimation](/docs/infrastructure-cost-estimation) and [What You Pay For](/docs/billing/what-you-pay-for).

## What creators still own

- The workflow graph, prompts, and tool configuration
- Product page copy, pricing margin, and input envelope
- Keys in API Vault when using BYOK / non-hosted credential modes
- Compliance of the product with [Creator Terms](/docs/creator-terms) and [Creator Guidelines](/docs/creator-guidelines)

## Related docs

- [Workflow Run Policy](/docs/workflow-run-policy)
- [Workflow Studio](/docs/builder/workflow-studio)
- [Sell to Humans and Agents](/docs/sell-everywhere)
- [API Create Run](/docs/api/create-run)

## Start building

Open [Workflow Studio](/builder) and publish a workflow. Edgaze will operate execution, progress, and settlement for every marketplace and API run.
