AI Agents

Use Velora as an execution layer for AI agents: quote, route, sign, submit, and recover with deterministic API flows.

AI agents are good at orchestration, but bad at guessing financial state. Velora gives agents a small set of deterministic trading primitives: quote, choose an execution path, build the action, collect the right signature, submit, and track status.

Use this section when you want Claude, Cursor, Codex, an MCP client, or your own agent runtime to build or operate a Velora integration.

What agents can do with Velora

Route user intent

Map user requests like “swap now”, “gasless swap”, or “protect me from MEV” to the right Velora flow.

Run the trade loop

Quote, build, sign, submit, and poll swaps, limit orders, and TWAPs using deterministic sequences.

Generate integrations

Copy verified prompts for Claude Code, Cursor, Codex, and other coding agents.

Connect over MCP

Point any MCP client at the hosted Velora server for quotes, guidance, and unsigned transactions.

Why Velora fits agents

Agents need APIs that return structured decisions, not ambiguous text. Velora exposes execution paths as explicit response objects:

• Top-level delta means continue the Delta order flow.
• Top-level market means continue the Market transaction flow.
• If neither is present, stop and ask the user to change the trade.

That lets an agent branch deterministically instead of guessing which route was selected.

Execution paths

Path	Use when	User signs	Agent tracks
Delta Swap	The user wants gasless execution, MEV protection, or solver competition for an immediate swap.	EIP-712 typed data	Order ID and status
Delta Limit Order	The user wants to trade only at a target price or better. This is a Delta LIMIT order built and submitted through the Delta API, not AugustusRFQ.	EIP-712 typed data	Order ID and fill status
Delta TWAP	The user wants one signed order that splits execution across time. Sell TWAPs preserve source amount; buy TWAPs preserve destination amount and cap source spend.	EIP-712 typed data	Slice execution and order status
Market API	The user wants immediate on-chain settlement.	Transaction	Transaction hash
AugustusRFQ	The user wants OTC/RFQ settlement with a counterparty.	RFQ order or transaction, depending on flow	RFQ fill state
Widget / SDK	The user wants an app integration, not a raw API flow.	Depends on configured flow	App-level state

The agent rule

Agents should not infer trading behavior from natural language alone. Always branch on the quote response shape: top-level delta means Delta, top-level market means Market.

Where to start

• Use Decision tables to route user intent and API responses.
• Use Five-call recipe to implement the canonical trade loop.
• Use Prompt library to generate working integrations with AI coding agents.
• Use Trading modes to understand mode=DELTA, mode=MARKET, and mode=ALL.
• Use Limit orders and TWAP when the user wants delayed execution, target prices, or time-sliced trades. For API-based agents, build these flows through /v2/delta/prices, /v2/delta/orders/build, /v2/delta/orders, and Delta order read endpoints.

Guardrails

Agents should:

• Never invent token addresses, chain IDs, quotes, transaction hashes, or order IDs.
• Ask before changing chain, recipient, spend direction, or execution mode.
• Ask for confirmation before submitting a signed order or sending a transaction.
• Requote when the quote expires, validation fails, or balances materially change.
• Treat delta and market as mutually exclusive top-level response paths.