Overview

A conversation process is a plan for the reasoning engine. A compound action is where integration logic lives.

Compound actions and conversational processes are foundational building blocks for creating agents that automate tasks and interact with users. Understanding when to use each helps developers design solutions tailored to trigger types and interaction needs.

• Compound Actions: These are designed for system-level triggers, enabling proactive automation in response to backend events or schedules. They excel in non-interactive, background workflows within ambient agents.
• Conversational Processes: These are built for user-level triggers, facilitating reactive, interactive dialogues. They shine in conversational agents where user input, context, and consent are central.

The choice between them depends on the agent's context: ambient agents typically rely on compound actions for event-driven logic, while conversational agents use processes for user-initiated flows. However, they aren't mutually exclusive, integrations allow hybrid scenarios, such as triggering compound actions from conversational processes for asynchronous tasks or loops.

flowchart TD
    subgraph Plugin_Ecosystem["Context Guide"]
        subgraph Ambient_World["Ambient Agents"]
            A[Ambient Agent]
            B[System Triggers<br>]
            C[Compound Actions<br>]
            A --> B --> C
        end

        subgraph Conversational_World["Conversational Agents"]
            D[Conversational Agent]
            E[User Triggers<br>]
            F[Conversational Processes<br>]
            D --> E --> F
        end

        Start[Developer Decision: Build Agent]
        Start -->|Proactive? System Events?| Ambient_World
        Start -->|Reactive? User Input?| Conversational_World

        %% Ambient_World <-->|Interconnect: Async Workflows, Loops| Conversational_World
        Conversational_World -->|Interconnect: Async Workflows, Loops| Ambient_World
        Ambient_World -->|Need User Input?| Link[Connecting Ambient and Conversational Agents]
        %% Conversational_World -->|Need Background Tasks? Trigger Compound Actions| Link
    end

    style Ambient_World fill:#f9f,stroke:#333,stroke-width:2px
    style Conversational_World fill:#bbf,stroke:#333,stroke-width:2px

While both can include multiple activities, they serve fundamentally different purposes — and misusing them leads to fragile, underperforming agents.

Key Rule

Never execute two action-based activities in a row inside a conversational process without collecting a slot in between.

• If you're chaining actions without user input → move that logic to a compound action.
• Every action in a process exposes outputs to the reasoning engine and user — not ideal for internal plumbing.
• Chaining actions directly in a process bypasses optimization and hurts long-term performance.

Most of the time, this pattern indicates anti-pattern usage.

When to Use Compound Actions

Use compound actions when your workflow is triggered at the system level, without requiring immediate user interaction. They are the go-to for ambient agents, which monitor external signals proactively.

Scenarios

• Event-Driven Automation: Respond to backend events, like a database update or webhook.
• Scheduled Tasks: Run periodic jobs, such as daily reports or cleanups.
• Background Processing: Perform computations or integrations that don't need user input, like syncing data across systems.

Example

In an ambient agent monitoring inventory levels:

• Trigger: System event (stock below threshold).
• Action: Compound action fetches data, calculates reorder quantities, and updates an ERP system.

Advantages: Fast execution, no user dependency, scalable for high-volume tasks.

Limitations: Lacks user context; cannot directly solicit input.

When to Use Conversational Processes

Use conversational processes when triggers are user-initiated and require interactive guidance, validation, or consent. These are core to conversational agents.

Scenarios

• User Input Collection: Gather details via dialogues (e.g., troubleshooting a device issue step-by-step).
• Consent-Driven Actions: Require approvals, like authorizing a purchase or access request.
• Multi-Turn Interactions: Handle complex queries needing clarification (e.g., booking travel with preferences).

Example

In a conversational agent for HR support:

• Trigger: User query ("Request time off").
• Process:
  • Ask for dates → collect slot
  • Ask for reason → collect slot
  • Check policy → action activity (but only one!)
  • Route to manager → collect approval
  • Submit → trigger compound action for async processing

Advantages: Maintains context for natural conversations, validates inputs in real-time, enhances user experience.

Limitations: Reactive only; not suited for unattended automation.

Plugin Design Best Practice

Almost every plugin should expose one conversational process with a single activity.

Only use multiple activities in a process when:

• You need user guidance across turns
• There are complex dependencies (e.g., field A enables field B)
• You must enforce policy via validators

Otherwise: keep integration logic in compound actions.

Blending Compound Actions and Conversational Processes

While compound actions and conversational processes operate in distinct worlds, they can integrate for powerful hybrid agents. For instance:

• From a conversational process, trigger a compound action to run asynchronous workflows (e.g., after user input, kick off a long-running sync).
• Use loops in compound actions for iterative tasks, or notify from ambient to conversational for user involvement when a task completes or fails.

For detailed guidance on bridging these, including using notify for transitions, refer to our guide on Connecting Ambient Agents and Conversational Agents.

This blend enables flexible architectures, such as ambient agents prompting user actions or conversational flows delegating to background processing.

Summary: Decision Checklist