Sail powers Agent-Operated SMAs with Definitive Flash

“We built Sail so an agent can manage money without ever holding user keys, so the execution layer had to work the same way. Definitive is professional execution with no custody.”

— Alejandro Dopico, co-founder, Sail

Highlights

• Sail is an AI lab building Sailor: the open source harness that turns your coding agent into a money agent.
• Sail’s reference agent holds MAG 5 stock tokens on Robinhood Chain, weighted by market cap and rebalanced daily with no human in the loop.
• Every trade executes through Definitive’s Flash API, giving Sail professional-grade routing without giving up custody or control.
• Flash ships as an MCP server, letting the same coding agent that deploys the SMA and writes its permissions wire in execution directly.
• One person, one evening: Sail went from an already-scaffolded SMA to live trades in a single coding session.
• Every routed swap is still validated onchain against the user’s mandate, including limits on tokens, trade size, approvals, and slippage.
• With execution solved, Sail can focus on new permission templates, agent skills, and forkable reference strategies rather than building and maintaining trading infrastructure chain by chain.

What is Sail?

Sail is an AI lab building Sailor: the open source harness that turns your coding agent into a money agent. Sailor runs inside Claude Code, Cursor, Codex, Hermes, or another coding agent. It deploys your account, writes the permissions that bound it, and runs your strategy, all on your own machine.

Your Sail account is a separately managed account (SMA): a Safe holding your capital that the agent can operate but never own. What the agent may do is defined by Solidity permission contracts, and the full set of them is your mandate.

Underneath sits Sail Protocol, the onchain rails. Every transaction is checked against your mandate before it executes. Anything outside the rules does not go through.

Sail’s thesis is simple: intelligence is becoming abundant. The harder problem is giving that intelligence access to real capital without giving it unrestricted control.

But permissions only solve half the problem. An agent also needs execution infrastructure that can reliably act within those constraints without sacrificing price, security, or custody.

Sail needed professional execution without compromising the mandate

Sail’s first live reference agent manages a portfolio of five tokenized stocks, rebalancing daily to stay aligned with market-cap weights.

For a strategy that trades every day, execution quality compounds. Poor routing or unnecessary slippage can erode the benefit of automating the strategy in the first place.

Building that infrastructure internally would mean taking on the same problem every trading product eventually faces: liquidity aggregation, routing, venue integrations, execution logic, and MEV protection.

And Sail operates across 10 chains, creating the risk of solving that problem repeatedly as it expands.

Custody was non-negotiable. Sail is designed so the agent never owns or controls the user’s funds. Any execution layer that required assets to leave the SMA’s custody would undermine that model.

The integration also needed to match how Sailor itself works: through a coding agent, not a traditional trading dashboard.

Definitive Flash: non-custodial execution built for agents

Definitive Flash exposes the execution engine behind Definitive’s institutional trading terminal through an AI-native API developers and agents can build on.

Built by former Coinbase Institutional trading engineers, Flash gives Sail a single execution layer for routing trades while leaving custody and policy enforcement with the user’s SMA.

Execution that stays inside the mandate

Flash never takes custody of the assets it routes.

Definitive determines how a trade should execute; Sail determines whether that trade is allowed to execute at all.

Sail’s permission contracts enforce those boundaries onchain. For example:

• RebalancerSwapPermission constrains the tokens, trade size, and slippage an agent can use.
• BoundedRouterApprove limits how much any individual spender can be approved to access.

Assets remain in the user-controlled account until execution conditions are met, even for conditional order types. When the agent submits a trade, Flash finds the route and Sail’s contracts independently verify that the transaction complies with the mandate.

Definitive optimizes execution. Sail enforces intent. Neither requires the user to surrender custody.

An MCP server that drops directly into the agent workflow

Flash ships as an MCP server, so Sail did not need to build a separate integration layer between its coding agent and its execution provider.

From the same session used to deploy an SMA and configure its permissions, the agent can add Flash with a single command: npx -y @definitive-fi/flash-mcp setup

The coding agent can then wire Flash directly into the strategy’s execution loop.

That made the integration unusually fast: one engineer moved from an already-scaffolded SMA to live trades on Robinhood Chain in one evening.

One execution integration across Sail’s multi-chain footprint

Sail supports 10 chains. Flash supports 11 EVM chains plus Solana, allowing one integration to cover most of Sail’s current surface area.

Instead of building routing infrastructure separately for each chain, Sail can integrate once and rely on Definitive to expand and maintain the execution layer underneath it.

As Flash adds new venues and routes, Sail can benefit without rebuilding its trading stack.

More time spent building Sail

Execution infrastructure is necessary, but it is not the product Sail wants to spend its engineering cycles reinventing.

With Flash handling routing, Sail can focus on the differentiated layer above it:

• New permission templates
• Additional Sailor skills
• More reference agents that developers can fork and customize.

Each new agent can inherit the same execution infrastructure instead of starting with another routing project.

Real impact

Sail’s MAG 5 reference agent is live on Robinhood Chain with capital deployed, automatically rebalancing its portfolio every day with no human in the loop.

The integration path was equally important:

One person. One evening. One coding session from an already-scaffolded SMA to live execution through Flash.

And every transaction still passes through Sail’s onchain mandate before it can execute.

The result is an agent that can autonomously manage real capital while neither the agent nor its execution provider takes custody of the funds.

What’s next: infrastructure for agent-managed capital

Sail’s bet is that the bottleneck in agentic finance is shifting.

The question is no longer simply whether an agent can make an investment decision. It is whether that agent can be trusted to act on real money within boundaries the user defines and the agent cannot bypass.

Sail provides those boundaries.

Definitive provides the execution layer underneath them.

Together, they create a reusable architecture for agent-managed capital: onchain mandates that constrain what an agent can do, paired with professional-grade, non-custodial execution that lets the agent actually do it.

For teams building agent-operated accounts, that changes the implementation problem from building a trading stack to defining the strategy and its permissions.

As Sail expands its library of permission templates, skills, and reference agents across its supported chains, those agents can inherit the same execution rails without a new integration project each time.

Definitive’s mission is to bring professional-grade execution wherever apps and agents need to trade.

Sail shows what that looks like when the trader is software, the mandate is enforced onchain, and the user is in full control.

Learn more about Sail: sail.money | Docs: docs.sail.money | X: @SaildotMoney

Learn more about Definitive's Flash API: flash.definitive.fi | X: @definitivefi