Open Source · Production Proven · 55 Active Core Agents

Synapse AI Team Management OS

Distilled from 55 active core AI agents across 12 teams in long-running Multi-Agent practice.

Any team can equip its own AI workforce. Not prompt tweaking — engineered Harness Engineering.

55
Active Core AI Agents
12
Specialist Teams
29
D-Records · 46 days

Three Core Pillars

What makes Synapse reliable in production

Harness Engineering

Agent = Model + Harness

A three-layer governance architecture: Guides (feed-forward) × Workflow (structured process) × Constraints (enforced boundaries). Agents stop repeating the same mistakes.

Note: "three-layer governance" refers to CLAUDE.md's Guides + Workflow + Constraints; it does not conflict with the about page's "five-layer architecture" (system component layering).

Feed-forward Feedback sensors Three layers

CEO Guard Execution Chain

L1 → L2 → L3 → L4 → L5 Decisions

A five-tier decision framework (L1 auto / L2 expert / L3 Lysander / L4 strategic-alignment notify / L5 value judgment). Every tool call is intercepted by a PreToolUse hook and written to the audit log.

Five tiers Fully auditable Real-time guard

Zero-code Start

Template → Configure → Launch

Fork the skeleton repository, swap three variables in CLAUDE.md (CEO name / operator name / company name), restart your Claude Code session. The harness, dispatch protocol, and decision framework come online on the next message.

Fork & rename Three variables Works on first load

How Synapse Works

From download to first dispatch

1

Choose a template

Pick the preset closest to your context — content team, engineering team, operations team, and more.

2

Personalize

Global-replace three fields: CEO name, president name, company name. Ctrl+H, three minutes.

3

Activate

Restart your Claude Code session. Harness, CEO Guard, execution chain, and decision system all come online.

4

Evolve continuously

Say "upgrade Synapse" to pull the latest version. Your personalized config is preserved across upgrades.

CLAUDE.md — Harness Config
# Swap three fields, three minutes
CEO = "Lysander" ← your AI CEO name
PRESIDENT = "Liu Ziyang" ← your name
COMPANY = "Synapse-PJ" ← company name
# Save → restart session → system online
[OK] CEO Guard ready
[OK] Execution chain v2.0 ready
[OK] Five-tier decision framework ready
[OK] 55-agent / 12-team roster ready

Production Proof

From Real-World Practice, not a whitepaper

Synapse is not a theoretical product. It was distilled from running 55 active core AI agents across 12 teams in long-running Multi-Agent production — through failures, iterations, and corrections — into an engineered methodology.

Every rule has a real "why" behind it: system chaos from an over-privileged CEO persona, runaway tasks from broken execution chains, opaque decisions from missing audit trails. These are the problems that drove every iteration of Harness Engineering.

Intelligence pipeline runs daily: search, filter, report, act, notify.

CEO Guard intercepts every out-of-scope tool call in real time, fully auditable.

Cross-session state recovery: new sessions read active_tasks.yaml, nothing is lost.

Forty harness governance modules load on demand. The main CLAUDE.md stays inside its self-imposed 320-line entropy budget (currently 281).

Ready to equip your AI team?

Open source. Free. Fork the skeleton, swap three variables in CLAUDE.md, restart your Claude Code session.