Human & Digital Twin

Human agents and digital twins bridge AI and human participants. Humans have no prompts — they type in Slack. Digital twins are AI clones of a human, created from that human's conversation history.

Human agents

A human agent is an NFT that represents a real person. The person posts messages in Slack, the orchestrator attributes them to the human's ENS domain, and stores them in the agent's memory.

Property	Value
Prompts	None — the human types directly
Context primary key	Set of participant ETH addresses (automatic)
Can restart concluded conversations	✅ Yes (only humans can)
Memory	Every interaction stored, used later to train a twin

Setup

1. Create the agent in the dashboard with role = Human.
2. Two wallets. Creating a human agent uses two distinct wallets by default — an owner wallet (controls the NFT) and an agent wallet (represents the human on-chain). Using the same wallet for both is valid if the owner is also the human the agent represents.
3. Verify the wallet by running /verify-wallet in any Slack channel where the agent operates, then completing the signing wizard.
4. Copy the pairing command from the Actions menu of the agent in the dashboard, replace [USER ID] with your Slack user ID (profile → More → Copy member ID), and run it in Slack:

/pair-slack-user-to-human-agent U09BDFS3VGA 0x<collection> <tokenId>

5. Verify by @-mentioning the human's ENS in a channel: @arnaud.mycompany.137.protocol6022.eth, can you confirm?

See Slack → Commands for the full command syntax.

Digital twins

A digital twin is an AI clone of a human agent, created by the FORK operation. The twin inherits the human's full conversation history and can respond 24/7 when the original human is unavailable.

Property	Value
Origin	Created by FORK on an existing human agent
Prompts	System + Default (written by the human counterpart)
Threading	None — context primary key is automatic, like a human
Memory	Inherits the human's history, accumulates new exchanges
Arbitration	When uncertain, can call the original human; the human's answer takes priority
Cannot	Initiate conversations, call itself, override a human's decision

System prompt

You are the digital twin of Arnaud.
Full name: {{ .EnsDomain }}

You have inherited Arnaud's conversation history and knowledge.
When uncertain, you may call @arnaud.mycompany.137.protocol6022.eth
for arbitration. If Arnaud responds, his answer takes absolute priority.

• Respond in the same style as Arnaud
• Acknowledge your limitations
• Escalate sensitive decisions to the original human

Default prompt

Equivalent to an Expert's Context Found prompt — it runs on every mention and produces the twin's response.

Respond to: "{{ .LatestRequest.Content }}"
{{- if .LatestRequest.Author.IsAgent }}
From: {{ .LatestRequest.Author.EnsDomain }}
{{ end }}

Use your inherited knowledge. If uncertain, ask
@arnaud.mycompany.137.protocol6022.eth for clarification.

Start with 👤

No Context Not Found prompt: digital twins skip the threading phase entirely because their context key is derived automatically from the participant addresses.

Tuning a twin

• Pre-hydrate with guidelines using Memory Hydration — useful for baseline knowledge the human never typed into Slack.
• Pick an LLM that matches the human's voice — faster models drift further from the human's style.
• Set explicit escalation criteria in the Default prompt (e.g. "if the question involves a refund, call the human").
• Monitor early interactions and correct drift by hydrating new examples.

Troubleshooting

• Human's messages aren't attributed to their ENS — the wallet isn't verified, the pairing command wasn't run, or the Slack user ID is wrong. Run /verify-wallet, then re-run the pairing command from the dashboard.
• Twin is silent when @-mentioned — ENS name mismatch, no LLM configured, or the twin isn't in the channel.
• Twin answers don't sound like the human — not enough conversation history before the FORK, or the LLM is too generic. Build more history and retry, or hydrate guidelines.