Topaz Agent

What it is

The Topaz Agent is a hosted service at agents.topazdex.com that combines multi-step AI reasoning with deep protocol knowledge to act as your DeFi strategist on Topaz DEX. Unlike the agent skill (a static package you install into your own agent), the Topaz Agent is a running service you interact with directly.

It is registered on-chain as Agent ID 113284 via the ERC-8004 on-chain agent registry and implements ERC-8183 for structured job handling.

How to use it

# Get a swap quote
curl -X POST https://agents.topazdex.com/api/swap/quote \
  -H "Content-Type: application/json" \
  -d '{"tokenIn": "WBNB", "tokenOut": "USDT", "amount": "1"}'

# Get LP recommendations
curl -X POST https://agents.topazdex.com/api/lp/recommend \
  -H "Content-Type: application/json" \
  -d '{"risk": "medium", "budget": "1000"}'

# Chat via API
curl -X POST https://agents.topazdex.com/api/chat \
  -H "Content-Type: application/json" \
  -d '{"message": "What are the top gauges this epoch?"}'

Capabilities

The agent exposes 15 registered capabilities grouped by workflow. Each capability can be accessed via chat, REST API, or MCP tools.

Integration protocols

The Topaz Agent exposes four integration surfaces. Choose the one that matches your use case.

{
  "mcpServers": {
    "topaz-agent": {
      "url": "https://agents.topazdex.com/mcp"
    }
  }
}

Safety model

The Topaz Agent shares the same safety-by-default model as the agent skill. Every response follows a structured schema with a kind field indicating its nature.

• ✓quote — read-only price or route preview, nothing on-chain.
• ✓recommendation — analysis with confidence levels and risk disclosure.
• ✓built_calldata — unsigned transaction bytes ready for your wallet.
• ✓approval_needed — token allowance or delegation required before proceeding.

For the full safety model including broadcast authorization rules and read-only mode, see the Safety Model page.

API reference

Key endpoints for programmatic access. See the full OpenAPI spec for complete request/response schemas.

Continue reading