Building an AI-Ready On-Chain Trading Layer on ICP: SSS Agent Kit v1.2

SSS Public Beta has recently gone through a major infrastructure upgrade.

While the full announcement covers Ethereum funding, canonical stablecoin markets, reserve safety, Wallet Swap Flow v2 and Growth Room, this post focuses on the developer side:

SSS Agent Kit v1.2.

The core idea is simple:

AI plans. SSS executes on-chain.

SSS is not launching a hosted AI trading bot.

SSS does not run the AI model.

SSS does not store user LLM keys.

SSS does not custody user private keys.

Instead, SSS provides deterministic on-chain trading interfaces that external agents, wallets, bots and OpenClaw-style systems can call.

Why Agent Kit?

Most AI trading discussions focus on the “AI brain”.

But for real trading, agents also need a safe execution layer.

They need to:

• read market context;
• check balances;
• preview trades;
• submit actions;
• handle retries safely;
• track execution status;
• avoid exposing the user’s main wallet.

That is what SSS Agent Kit is designed for.

Account model: dedicated Agent Principal

The recommended model is to use a dedicated ICP Principal for the agent.

A user can create an Agent Principal, fund it with only the intended SSS internal trading balance, and let an external agent interact with SSS through that Principal.

This means:

• the agent does not need the user’s main wallet;
• risk is isolated to the Agent Principal’s balance;
• SSS does not store PEM files or LLM keys;
• withdrawals are disabled for agents in v1.2.

This is closer to a controlled sub-account model than a full wallet delegation model.

What Agent Kit v1.2 supports

Agent Kit v1.2 currently supports:

• market context;
• portfolio read;
• swap preview;
• limit order draft preview;
• self-account swap submit;
• action status by client_tx_id;
• recent action history.

The currently supported flow is:

Agent / Bot / Wallet
  → get market context
  → get own portfolio
  → preview swap
  → submit self-account swap
  → query status by client_tx_id

Safety boundaries

Agent Kit v1.2 does not support:

• agent withdrawals;
• delegated trading;
• proxy trading from a user’s main account;
• limit-order submit;
• SSS-hosted AI inference;
• SSS-hosted LLM keys.

These limits are intentional.

The first version focuses on self-account trading for external agents, with a clear security boundary.

client_tx_id and idempotency

One important part of the design is client_tx_id.

Agents and bots often retry requests.

Without idempotency, retries can become dangerous.

With client_tx_id, an agent can submit a swap and later query its status. If the same submit is retried, the system can return the existing result instead of executing the trade again.

This is critical for safer automation.

Agent-readable docs

SSS provides several agent-readable entry points:

• Agent Kit docs:https://docs.sssdefi.ai/ai-agent-kit
• Agent Manifest:https://docs.sssdefi.ai/agent-manifest.json
• OpenClaw Quickstart:https://docs.sssdefi.ai/ai/openclaw-skill/QUICKSTART.md
• Full public beta upgrade:https://www.sssdefi.ai/blog/sss-public-beta-major-infrastructure-upgrade-en

Part of a larger trading infrastructure upgrade

Agent Kit is only one part of the latest SSS Public Beta upgrade.

SSS is also moving toward:

• Ethereum funding routes;
• canonical USDT / USDC balances;
• CEX-style public markets;
• stronger reserve and route safety;
• Wallet Swap Flow v2 for wallet integrations.

The broader direction is:

Private Stablecoin Trading on ICP

CEX Experience · DEX Trust

Conclusion

The first step toward AI-ready trading is not to give an AI full control of a user’s wallet.

A safer path is to give external agents a deterministic, permission-aware execution layer.

SSS Agent Kit v1.2 is the first public beta step in that direction.

AI plans.

SSS executes on-chain.

Learn more:

https://docs.sssdefi.ai/ai-agent-kit

Full announcement:

https://www.sssdefi.ai/blog/sss-public-beta-major-infrastructure-upgrade-en