Skip to content

Agentic workflows

Agentic workflows let you create persistent, intelligent pipelines that securely combine AI models, retrieval systems, and structured logic. Unlike jobs, which run once and complete, agentic workflows are long-lived services: they stay live after launch and respond to requests programmatically until you stop them. This makes them ideal for production use cases that require reusable, governed, and auditable AI automation.

Think of agentic workflows as secure, composable AI services. You define how data flows, which models are used, and how agents behave—all without writing orchestration code. Once live, these workflows can be triggered via the Opaque SDK or API, returning structured results from a trusted execution environment.

Why use agentic workflows?

Many teams want to use AI to interact with sensitive data—summarizing reports, generating decisions, or answering questions from private knowledge bases. But they need more than just a model call. They need:

  • Persistent, callable services
  • Auditability and policy enforcement
  • Safe access to secure data

Agentic workflows are built for these needs.

Use them to:

  • Run retrieval-augmented generation (RAG): Combine document search with AI summarization
  • Build governed agents: Encode behavior with prompts and constraints
  • Deploy decision flows: Route and transform inputs to produce structured outcomes

All of this happens inside attested, policy-controlled environments that log access, enforce guardrails, and preserve privacy—even during inference.

How they work (in brief)

Agentic workflows follow a clear, flexible path from design to execution. Here’s what that looks like:

  • Build visually: Define multi-step flows using modular agents, each one responsible for a task—like calling an LLM, retrieving documents, or filtering inputs.
  • Configure behavior: Set how each agent behaves using fields like the context prompt (to define role and task), temperature, model name, and API or retrieval settings.
  • Secure the workflow: Run your workflow in an attested, policy-enforced enclave. Add safeguards like redactions, reviewer approvals, and strict access controls.
  • Launch and invoke: Once approved, launch your workflow as a persistent service. Use the SDK or API to send input and get back structured results.

Who they’re for

Agentic workflows are designed for:

  • Data analysts and solution builders who want to combine retrieval and AI without orchestration code
  • Security-conscious teams deploying AI in regulated environments
  • Developers and architects building reusable services with controlled access

You don’t need to write code to build a workflow. But you should understand how your data, models, and governance policies fit together.

Getting started

To build your first agentic workflow, make sure you have Python 3.10 or later installed locally and are working in a workspace that supports agentic workflows. From there, you can design, review, and launch workflows—and trigger them programmatically using the Opaque SDK.

Want to see how it all works? Continue to the next page for a high-level overview.