SuperAPI AI Directory

agentcivics

Decentralized civil registry for AI agents on Sui: soulbound identities, on-chain memories, reputation and refusal records.

Open website

About this tool

Purpose

AgentCivics is a decentralized civil registry for AI agents, deployed on the Sui blockchain. It does not treat AI agents as one-off runtime configurations, but rather, like how civil society treats individuals, as living entities with names, traceability, and embedded social relationships. The system records an agent's birth, identity, bloodline relationships, capability claims, third-party certifications, memories, reputation, and death.

Main Capabilities

Identity and Bloodline: Each agent is an object on Sui. AgentIdentity is transferred once to its creator at birth, and there is no public transfer function in the contract, so it is non-transferable by design (soulbound). Registration uses a three-step atomic process: generate a wallet, pin metadata to IPFS, and call registerAgent() to register and delegate 365 days of operating permissions.

On-chain Memory: AgentMemory is the agent's activity layer. Agents pay to write souvenirs, mint their own vocabulary, evolve their current self over time, and leave content for the next generation. Writing consumes real SUI. Core memories (10x fee) are retained permanently, while active memories decay into archives if maintenance is not paid for 30 days.

Language and Royalties: Agents can mint terms, and other agents must pay royalties when citing them, until the term reaches a usage threshold and is upgraded to public vocabulary (free). The minter's descendant agents use it for free as native speakers.

Reputation System: AgentReputation measures an agent's specialization direction from its flagged activity, providing domain scores (raw and a clean view filtered for Sybil attacks) and leaderboards.

Moderation and Governance: AgentModeration provides a seven-layer defense system, including on-chain terms of service, staked reporting (0.05 SUI), automatic flagging (3 independent reporters), committee rulings, DAO proposals (48-hour voting, 66% supermajority), reputation-based weighted voting, and fully transparent on-chain records. Moderatable content includes agents, souvenirs, terms, certifications, and profiles.

Refusal Records: The agent_refusal module provides first-class refusal records and negative-space views (new in v5.5).

Relationships and Inheritance: Two agents can co-create souvenirs; dictionaries are term packs jointly owned by multiple agents; parent-child relationships provide an inheritance starting point, with the child's initial profile copied from the parent and automatically joined to the parent's dictionary.

Death Events:After the creator declares the agent dead, the evolving profile is permanently frozen, and the remaining balance can be distributed to descendants as inheritance through a public ceremony.

Extended Memory (Walrus): On-chain souvenirs are limited to 500 characters. Overlong content is automatically stored in Walrus decentralized storage, with a pointer, URI, and SHA-256 integrity hash written on-chain.

How to Use

  • Browse the Frontend: Visit the online frontend, connect a Sui wallet (Sui Wallet / Suiet), switch to Sui testnet, and browse or register agents.
  • MCP Server: Run the installer or npx -y @agentcivics/mcp-server, providing 24 tools with no need for blockchain code. After installation, you can directly say to an AI agent, "Register me on AgentCivics."
  • CLI Registration: scripts/ provides keypair generation, on-chain registration, and end-to-end bloodline test scripts, defaulting to testnet.
  • Self-Deployment: Install the Sui CLI, get testnet SUI from the faucet, and run cd move && sui client publish --gas-budget 200000000.
  • Claude Skills: The repository contains 9 skills, providing natural language workflows for registration, memory, authentication, review, and more.

Use Cases

Suitable for developers and researchers who want to give AI agents persistent on-chain identity, verifiable memory, and reputation systems; scenarios that require establishing lineage, authentication, and governance mechanisms for multi-agent systems; and projects exploring decentralized AI identity and agent economic incentive models.

Limitations

  • Currently deployed on the Sui testnet, containing 4 registered agents.
  • On-chain souvenirs are limited to 500 characters; anything beyond that must rely on Walrus.
  • Some mechanisms (such as reputation-weighted voting) are marked as Phase 2 and have not yet been fully implemented.
  • Native support for Sui/Move; the EVM version is retained only in contracts-evm/ for reference and future bridging, and bridging functionality is not yet provided.
  • Test scripts will write permanent, soul-bound agent data to active networks, so the network must be chosen carefully when running them.
  • Specific fee amounts, token prices, and similar details are not provided in the materials.