From input to output — watch how information flows through the agent's brain, gets remembered, and loops until the goal is done.
Every component belongs to one of three jobs: getting information in, processing it, or getting results out. Hover each card for details.
Takes in your request, documents, data streams, and live signals — everything the agent needs to understand the situation.
The reasoning engine — reads the goal, weighs the options, and decides the next move.
Breaks big goals into ordered steps, reflects on results, and re-plans when a step fails.
Short-term memory keeps the task in focus; long-term memory recalls facts and past lessons.
Web search, code execution, databases, and APIs — reached through standards like MCP.
Decisions become real effects — a written answer, an edited file, an email sent, a workflow triggered.
This is the heartbeat of every agent. Information flows into the brain, which consults memory, then runs the execution branch — plan the step, pick the tool, act — and the result loops back to the brain until the goal is complete.
The agent perceives: your request comes in along with files, data, and live context from the environment. This is everything the brain will reason about.
The LLM interprets the input and pulls context from memory. Then it kicks off the execution branch: plan the step, pick the tool, act.
A two-way street with the brain: it reads past context before deciding, and writes new results after acting — so every loop is smarter than the last.
The brain decides the next concrete step: what needs to happen now, and what ability is required to do it.
For the planned step, the agent selects the right ability: web search, code execution, a database query, or an API call through MCP.
The tool runs and the result lands in the real world: an answer, an edited file, an email sent. The outcome loops back to the brain to be observed.
That's the agent process — the brain thinks with memory at its side, the execution branch does the work, and the loop runs until the goal is achieved.