Introduction to Multi-Agent Architecture
Complex tasks often require continuous attention or deep focus on a narrow aspect of a problem. If Mother AI tries to do everything sequentially, it can lose track of the main goal or waste context window on minor details.
Flowork solves this through a Multi-Agent Spawning System.
How Mother AI Uses Agents
When faced with a tedious or specialized task, Mother AI acts as an orchestrator.
For example, if the user asks, "Test the entire login flow on this staging site."
1. Mother AI spawns a browser_agent using the spawn_agent tool, giving it specific instructions: "Navigate to URL, input test credentials, submit, and report success or failure."
2. Mother AI continues planning or communicating with the user.
3. The browser_agent uses its own dedicated toolset (tailored for browser interaction) to complete the task.
4. Once finished, it returns a concise summary of its actions and the outcome to Mother AI.
Agent Types
- Browser Agent: Specialized in visual interaction. It can view screenshots, click elements, and inject text, abstracting the complex raw DOM manipulation from the main AI.
- Monitor Agent: Specialized in observing streams of data, such as a running server log, and alerting the main AI only when a specific condition (like an error or a "server started" message) occurs.