Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.dzap.io/llms.txt

Use this file to discover all available pages before exploring further.

Vibe coding = describe outcomes, not steps. The agent picks tools. You sanity-check.

​
Patterns that work

​
1. Anchor on accounts

The agent does better when wallet + chain are in metadata, not buried in the prompt: 
await sdk.ask({
  userQuery: "What's my exposure to ETH?",
  metadata: {
    accountInfo: [
      { blockchain: 'evm', chain: '1', user_account: '0xUser' },
      { blockchain: 'evm', chain: '42161', user_account: '0xUser' },
    ],
  },
});

​
2. Ask for plans before actions

For trades, two prompts beats one: 
1. "Plan a swap of 100 USDC to WETH on Arbitrum. Show route and fees."
2. "OK, execute it."
The first surfaces a route for review; the second commits. Keeps you in the loop without sacrificing speed.

​
3. Composing tools

Don’t manually call tools in series — the agent’s better at it: 
"Show top 3 pools on Base, pick the highest APR, plan a zap from USDC."
ZapBot will fire PoolTool → reason about APR → fire ZapCallDataGeneratorTool.

​
4. Constraints in plain English

"Swap 100 USDC for the best stablecoin yield on Arbitrum, max 1% slippage, no protocols below $10M TVL."
The agent passes constraints as parameters where supported and rejects routes where not.

​
Gotchas

​
Hallucinated addresses

LLMs can invent token addresses. Always pass an account or token via metadata when accuracy matters; fall back to TokenAddressTool to look up by symbol.

​
Slippage units

Tools and APIs use percent, not basis points. 1 = 1%. The agent gets this right when you say “1%”; it sometimes converts oddly when you say “100 bps”.

​
Native tokens

Native ETH/MATIC/etc. uses the sentinel 0xEeeeeEeeeEeEeeEeEeEeeEEEeeeeEeeeeeeeEEeE. Tools handle this transparently when you say “ETH”; they get confused if you pass 0x0...0 (zero address).

​
Multi-chain ambiguity

“Swap USDC for WETH” without a chain is ambiguous. Either set chain in metadata or specify (“on Arbitrum”). The agent will ask for clarification if it’s missing.

​
Bridge timing

Cross-chain bridges aren’t instant. The agent kicks off getZapTxnStatus polling but a UI loop should also poll independently — bridges can settle in seconds (Across) or minutes (CCTP).

​
Recipes

​
Daily portfolio summary

"Summarize my portfolio across Ethereum, Arbitrum, and Base.
Group by token, show USD value, flag any positions worth less than $1 (dust)."

​
Rebalance to target

"My target: 60% USDC, 30% WETH, 10% WBTC across Arbitrum and Base.
Plan the swaps to hit target. Show me the plan; don't execute."

​
News-driven swap idea

"Show top 3 trending tokens on Base by 24h volume.
For the top one, pull recent news and sentiment.
If sentiment is positive, plan a $500 swap from USDC."

​
When NOT to use the agent

  • High-value trades — the agent’s good at planning; for $100k+ swaps, drive the SDK directly with explicit slippage and review the route.
  • Production wallets — never give a production hot wallet’s key to an unsupervised agent. Use session keys and limits. See ZapBot Safety.
  • Deterministic flows — if you know exactly what should happen every time, the SDK is faster and cheaper.