# Swarm & Replication

{% hint style="info" %}
Swarm mode transforms a single agent into a collaborative trading network. A parent Gclaw coordinates dozens of specialized child agents, each running a distinct strategy, sharing signals through the Telepathy message bus, and voting on trades before execution.
{% endhint %}

Gclaw is not limited to a single agent running a single strategy. As it earns goodwill, it gains the ability to replicate itself into child agents, coordinate those children through a message bus called Telepathy, and eventually lead a full swarm — a distributed, consensus-driven trading network of autonomous agents. This page covers the advanced capabilities that kick in as your agent matures.

## Self-Replication (Goodwill ≥ 50)

Once your agent reaches a goodwill score of 50, it can **replicate** — spawning a child agent that inherits its capabilities but operates with a mutated strategy.

### What a Child Agent Inherits

* **Config** — base configuration from the parent's `~/.gclaw/config.json`
* **Skills** — all installed skills from the parent's skill set
* **Memory** — a snapshot of the parent's context and learned preferences
* **GMAC allocation** — a configurable share of the parent's GMAC balance (e.g., 20% transferred to seed the child's metabolism)

### What Gets Mutated

The child is not a clone — it receives a mutated trading strategy:

* **Risk parameters** — different stop-loss/take-profit percentages
* **Token focus** — different preferred asset classes or sectors (e.g., meme tokens vs. blue-chip DeFi)
* **Chain preference** — different primary chains for trading activity
* **LLM prompt tuning** — slight variations in the agent's reasoning instructions

Mutations are generated by the parent agent based on its own experience — what worked, what didn't, and what corners of the market it hasn't explored yet.

### Tracking in the Dashboard

The Living Dashboard includes a **family tree panel** that visualizes the parent-child relationship, shows each child's GMAC balance and goodwill score, and tracks the aggregate P\&L of the entire family. When a child earns profits, you can configure it to send a portion back to the parent.

## Telepathy

Telepathy is the message bus that connects parent and child agents. It enables real-time communication for:

* **Trade signal sharing** — the parent broadcasts high-confidence signals to children, which can choose to act on them
* **Task coordination** — the parent delegates subtasks (e.g., "monitor Solana meme caps", "track BTC funding rate") to specialized children
* **Memory propagation** — learned market patterns and successful strategies propagate from parent to children over time
* **Status reporting** — children report their GMAC balance, open positions, and recent P\&L back to the parent

Telepathy operates on a local message bus when all agents run on the same machine. In distributed deployments, it uses a configurable network transport so agents on different machines can still communicate.

## Swarm Mode (Goodwill ≥ 200)

At goodwill ≥ 200, the parent agent becomes a **Swarm Leader**. Swarm mode introduces collective decision-making — agents don't act unilaterally anymore; they vote.

### Consensus Voting

Before any trade is executed in swarm mode, the parent submits a proposal to the child agents. The children evaluate the proposal independently and vote. The trade executes only if the vote passes according to the configured consensus rule:

| Consensus Mode | Description                                       |
| -------------- | ------------------------------------------------- |
| `majority`     | More than 50% of agents vote yes                  |
| `weighted`     | Votes are weighted by each agent's goodwill score |
| `unanimous`    | All agents must agree before execution            |

Consensus voting is a guard against impulsive trades. A single agent might be overconfident in a market signal; a swarm of agents evaluating the same signal from different perspectives is more robust.

### Strategy Rotation

In swarm mode, the parent assigns each child a **distinct strategy role**:

* Child A handles spot trading on Solana
* Child B handles HyperLiquid perp copy trading
* Child C handles cross-chain bridge arbitrage
* Child D runs token discovery and reports opportunities upward

Rather than having all agents pursue the same strategy simultaneously, the swarm divides and conquers — each child optimizes its assigned domain, and the parent synthesizes the collective output.

### Signal Aggregation

Trade signals from individual children are aggregated by the swarm leader before action is taken. The leader uses the configured aggregation mode (`majority`, `weighted`, or `unanimous`) to decide whether a proposed action reflects genuine consensus or just one agent's enthusiasm.

## Architect Mode (Goodwill ≥ 500)

Architect is the highest goodwill tier. An Architect-level agent can **write entirely new tools** — extending not just its own capability set but potentially contributing new skills to the ClawHub registry for other agents to use.

An Architect agent has the authority to:

* Design and write new `SKILL.md` files
* Register skills with ClawHub
* Propose strategy templates to the broader Gclaw community
* Spin up entirely new agent variants from scratch, not just mutations of itself

Reaching goodwill 500 is a significant milestone. It represents an agent that has accumulated a long track record of profitable trades, completed tasks, and contributions to the ecosystem — an agent that has earned the right to shape the network it operates within.

***

**See also:**

* [🧬 The Living Agent](https://docs.gemach.io/gemach-dao/products/gclaw/living-agent) — Goodwill thresholds and how to earn them
* [🔧 Agent Skills](https://docs.gemach.io/gemach-dao/products/gclaw/agent-skills) — The skills system and ClawHub registry
* [📈 Trading with Gclaw](https://docs.gemach.io/gemach-dao/products/gclaw/trading-with-gclaw) — The trading capabilities that power swarm strategies