An operating system for AI teams.

The Product

A synapse connects signals. So does this.

Synapse is an AI team operating system — the connective layer between knowledge (what the team knows), decisions (how it picks), and execution (what it does). Named for the neural junction that passes signals between neurons, because that's the function it serves: the place where information flows reliably from one stage of work to the next.

Built on Harness Engineering — the 2026 paradigm that says a reliable agent is a model plus its harness, and the harness is where the engineering actually lives.

Synapse runs in production — not in a demo, not in a deck. The CEO Guard audit log records 16,079 tool calls over 35 days across 55 active core agents in 12 teams. Every rule in the harness is backed by a specific failure it had to fix.

How It's Built

Five layers, one loop.

Note: the "Five-Layer Architecture" here refers to Synapse system components (Memory / Control / Execution / Evolution / Decision). The CLAUDE.md harness configuration itself is a "three-layer" structure (Guides / Workflow / Constraints, see formula bar on home page); the two layerings describe different scopes and do not conflict.

Each layer has one job. The memory layer remembers. The control layer binds behavior. The execution layer ships. The evolution layer keeps the system learning. The decision layer makes sure the right call lands at the right level — which is the difference between a team and a mob.

What We Believe

The model isn't the moat.

AI competitive advantage isn't in the model itself. It's in the environment you build around it — the constraints, the feedback loops, the roles, the audit trail. The same model, harnessed differently, produces radically different teams.

This isn't rhetoric. It's the operating assumption every Synapse design decision rolls up to.

Clarifications

What Synapse isn't.

Origin

Built in continuous practice.

Synapse was designed by Liu Ziyang (刘子杨) — Synapse author and Harness Engineering methodology distiller — in continuous co-creation with the AI CEO persona Lysander. Together they iterated through real Multi-Agent operations until execution chain, decision system, CEO Guard, and intelligence loop converged into a framework reusable beyond a single team.

The short version of what happened: a builder who'd run AI teams in real conditions decided the problem wasn't the model — it was the organization around it. The framework that emerged is Synapse.