The Brain Architecture Behind Wolffish
Why we modeled an AI agent after the brain
Most AI tools treat intelligence as a single function: input goes in, output comes out. But real intelligence doesn't work that way. The human brain is a system of systems — perception, memory, reasoning, and motor control all working together in a continuous loop.
Wolffish is built on that insight. Instead of a simple prompt-response pipeline, it uses a brain-inspired architecture where memory, reasoning, and action are separate but deeply integrated subsystems.
The three pillars
Memory
Wolffish maintains multiple memory layers, similar to how the brain separates working memory from long-term storage. Short-term context lives in the active conversation. Long-term knowledge persists in markdown files the agent can read and update over time.
Unlike traditional RAG systems that treat memory as a search problem, Wolffish's memory is structured and intentional. The agent knows what it knows — and what it doesn't.
Reasoning
The reasoning layer is where Wolffish decides what to do. It evaluates the current context, consults memory, and forms a plan. This isn't just chain-of-thought prompting — it's a structured decision loop that considers constraints, risks, and alternatives before acting.
Think of it as the prefrontal cortex: the part that stops you from doing the first thing that comes to mind and instead considers whether it's actually a good idea.
Action
The action layer is what makes Wolffish an agent rather than a chatbot. It can execute shell commands, manipulate files, interact with APIs, and control applications. Every action is logged, reversible where possible, and bounded by permissions you control.
The loop
These three systems form a continuous loop:
- Perceive — Read the current state (files, terminal output, user input)
- Remember — Retrieve relevant context from memory
- Reason — Decide on the best course of action
- Act — Execute the plan
- Learn — Update memory with new information
This loop runs continuously, allowing the agent to handle multi-step tasks that would be impossible for a simple chatbot. It can start a task, encounter an obstacle, reason about alternatives, try a different approach, and eventually succeed — all without human intervention.
Why markdown matters
Every layer of the brain architecture is backed by plain markdown files. This isn't a technical limitation — it's a design philosophy. Markdown is human-readable, version-controllable, and universally portable.
Your agent's memory? A folder of .md files you can open in any text editor. Its skills? More markdown. Its reasoning traces? Logged in markdown. You can literally read your agent's mind.
This transparency is what separates Wolffish from black-box AI systems. You don't just use the agent — you understand it.