Pipelines

An introduction to Pipelines, a UI-agnostic plugin framework that enables custom, scalable, and extensible AI workflows using intrallmai-compatible APIs.

Pipelines

Pipelines is a UI-agnostic plugin framework designed to extend intrallmai API–compatible clients with custom logic, advanced workflows, and scalable execution. It enables developers and platform operators to build modular pipelines that run independently from the main UI, allowing complex or resource-intensive tasks to be handled efficiently.

Pipelines integrate seamlessly with any client that supports intrallmai API specifications, making them suitable for flexible and distributed AI architectures.

When to Use Pipelines

Pipelines are not required for all extensions.

You should consider Pipelines when:

  • You need to run computationally heavy workloads
  • You want to offload execution from the main UI for performance or scalability
  • You are building custom workflows that go beyond simple request handling
  • You need tight control over execution logic using Python

If your goal is only to:

  • Add support for additional providers
  • Apply simple filters or transformations

Then built-in Functions are usually a simpler and more appropriate choice.

What Pipelines Enable

Pipelines allow you to:

  • Inject custom logic into request and response flows
  • Integrate any Python library or external system
  • Build reusable and composable workflows
  • Extend AI behavior without modifying the core UI

They act as an external execution layer while remaining fully compatible with intrallmai-style APIs.

Key Capabilities

Modular Workflows

Pipelines are composed of independent, reusable modules that can be combined to form complex workflows.

UI-Agnostic Integration

Any UI or client that supports intrallmai API specifications can leverage Pipelines by pointing to the Pipelines endpoint.

Custom Hooks

Developers can define custom hooks to intercept, transform, or augment requests and responses.

Python-Native Extensibility

Pipelines are written in Python, enabling direct use of:

  • AI frameworks
  • Databases
  • Monitoring tools
  • External APIs
  • Internal services

Example Use Cases

Pipelines can be used to implement:

  • Function-calling pipelines for advanced tool orchestration
  • Custom RAG pipelines with tailored retrieval and ranking logic
  • Message monitoring and observability integrations
  • Rate limiting and traffic control
  • Real-time translation and content transformation
  • Content moderation and filtering

These examples illustrate how Pipelines can handle logic that is too complex or resource-intensive for inline execution.

How Pipelines Fit Into the Platform

  • Tools and Functions define what capabilities are available
  • Pipelines define how requests are processed and executed
  • UI clients remain lightweight and focused on interaction

This separation improves scalability, maintainability, and performance.

Security Considerations

Pipelines execute arbitrary Python code. As such:

  • Only trusted pipelines should be deployed
  • Pipelines should be treated as privileged services
  • Access and network boundaries should be carefully controlled

Always review pipeline logic before enabling it in production environments.

Status and Roadmap

Pipelines are actively evolving, with ongoing improvements to:

  • Plugin hooks
  • Execution control
  • Observability
  • Developer experience

The long-term vision is to establish Pipelines as a flexible plugin ecosystem for AI interfaces, enabling rich extensions without coupling logic to the UI layer.

This page introduces Pipelines conceptually. For configuration and management, refer to the Pipelines settings and administration documentation.