Backend pipeline

Backend pipeline architecture

Build interactive AI applications without keeping requests open. Your users see live progress while your backend continues working in the background.

Compare architectures
1
Recommended

Backend pipeline

Live progress updates, background AI workflows, and backend processing. Best for interactive AI applications.

Learn more
2

Async (Dual-Delivery)

Submit once, then automatically deliver results to both your backend and your users. Best for reliable asynchronous processing.

Learn more
3

Sync

Wait for AI to finish before returning the result. Best for quick AI requests and straightforward integrations.

Learn more

Backend pipeline flow

See how ModelRiver coordinates AI and your backend from the first request to the final response.

Customer asks for a refund → AI understands the request → your backend checks the CRM → AI writes the final reply.

1
User prompt
Frontend interaction
2
Your backend
Sends AI request
3
ModelRiver
Coordinates AI requests
4
AI providers
AI generates a response
Background workflow starts
6
Your webhook
Check CRM, save data,
call APIs
5
ModelRiver
Continues the workflow
7
User's browser
User sees the update instantly
Architecture

Why backend pipeline?

Backend pipelines let AI keep working after your app responds. Your users don't have to wait, your backend stays responsive, and ModelRiver handles the heavy lifting behind the scenes.

Fast Acknowledgment

Return a response immediately while AI continues working in the background. Your users don't have to wait for long-running AI tasks to finish.

Background Execution

Run large AI tasks in the background instead of slowing down your backend. Your application stays responsive while ModelRiver manages the heavy AI work.

Automatic Fallback

If one AI provider becomes slow or unavailable, ModelRiver automatically switches to another so your workflows continue without interruption.

Live progress updates

Show progress and final results as soon as they're available. Users see updates instantly without refreshing the page.

Ready to start?

Start building View documentation