Flows

In Robility flow, flows represent the complete pipeline of how data moves and gets processed across connected components. A flow is essentially the visual workflow you design on the canvas by dragging, dropping, and linking components together. Each flow starts with an input (like a user query, file, or external request), passes through a sequence of core components (data processing, logic, model calls, or tool usage), and ends with an output that returns the final result.

Flows can be simple, such as connecting a prompt input directly to a language model, or highly complex, involving loops, conditional branches, external API calls, and agents for dynamic decision-making. They can be run interactively for testing, served through APIs, or embedded into applications—making them flexible for both experimentation and production use.

In short, flows are the orchestrated pathways that bring components, agents, and tools together into a working system that delivers intelligent, end-to-end automation. 

How Flows Work

1. Starting Point (Inputs) – Every flow begins with one or more input components. These can be direct user prompts, uploaded files, database queries, or external API calls. The input acts as the entry point of data into the workflow.

2. Processing Stage (Core Components) – Once the data enters the flow, it moves through a chain of core components. These may include:

a. Data Processing (splitting, cleaning, embedding text, structuring data).

b. Logic Control (routers, conditionals, or loops to handle branching scenarios).

c. Language Model Calls (passing prompts and context to an LLM for reasoning or generation).

3. Dynamic Decisions (Agents) – For more complex workflows, flows can integrate agents. Agents dynamically decide which tools or components to invoke, creating a layer of autonomy within the flow.

4. External Integrations (Tools & APIs) – Flows can call external services or APIs through tool components. For example, a flow could retrieve real-time data from a knowledge base, perform calculations, or connect to third-party systems like Google Sheets or databases.

5. Output Generation – Finally, the flow delivers results via output components. This could be a direct response in natural language, a structured dataset, a visualization, or even an automated action like sending an email or updating a record in a system.