Workflows

Workflows are the core logic components that define how an agent operates. Each workflow consists of four key elements:

1. A trigger that initiates the workflow

2. An AI model that processes the workflow's logic

3. Plugin functions that AI model can trigger

4. A set of instructions that guide the agent's actions

This structure makes workflows powerful and flexible. Rather than having a single, monolithic set of instructions for all agent behaviors, specialized workflows can be created for different scenarios.

Each workflow can use the most appropriate AI model and plugins for its specific task and can be triggered in various ways.

Workflow advantages

• Organized and maintainable agent logic

• Task optimization with specialized models

• Workflow reusability across different scenarios

• Simplified testing and debugging of specific agent behaviors

• Creation of complex agent behaviors from simple building blocks

Workflow Configuration

Workflow Triggers

Triggers can be automatic events like scheduled intervals (running every 10 minutes), or Plugin triggers (receiving mentions on X/Twitter).

Custom workflows can also exist without automatic triggers, allowing them to be called directly by other workflows.

Plugin access control

While agents can access all plugins by default within a workflow, you can restrict plugin access to improve performance and security. For example, you might want to prevent a workflow triggered by Twitter replies from accessing wallet functions. By limiting plugins to only those necessary for the specific task, you can optimize the context and improve the agent's accuracy.

Model selection

Each workflow can use a different AI model based on your needs. Choose from your installed models in the main builder page to optimize performance for specific tasks - whether that's summarizing content, analyzing images, or executing plugin functions.

Example

Let's look at how this works with a Twitter monitoring agent:

The main workflow orchestrates checking the Twitter feed and analyzing new tokens. Since social media feeds contain large amounts of content that can exceed AI model processing limits, the main workflow can trigger a separate "summarizer" workflow using the Gemini model, which handles large content effectively. The Gemini workflow returns a summary to the main workflow, which can then perform token analysis or execute trades.

This modular approach allows each component to focus on its strengths while working together efficiently.