Your users have a Visa card. A yield-bearing dollar account. Instant off-ramps to 65+ countries and 14,000+ banks. Native Bitcoin. Cross-chain transfers to Ethereum, Polygon, Base, and Solana. All inside your app, with Lightspark handling licensing, compliance, card issuance, and the payout rails.

That's what Grid Global Accounts gives them today. Now it gives those same capabilities to their AI agents.

AI agents as account delegates

AI agents are getting better at taking real action in the world. They can book meetings, write code, file taxes, and search the web. But when it comes to money, they still hit a wall. They can reason about finances. They can't actually move funds without a human stepping in.

Grid Global Accounts changes that. Account holders can now create scoped, revocable, auditable operating pockets for their AI agents, with spending limits, approved payees, and approval thresholds they define. Grid evaluates policy before any agent-initiated action executes. The user's funds stay theirs. The agent just gets to work.

For partners, the integration story is simple: the same API that gives your users a global financial account also makes it agent-ready. You surface the connection flow, the policy configuration screen, and the activity history. Grid manages the authorization layer underneath.

What users see

In a partner-built app, agent management sits alongside the user's account home. Users see their balance, their active agents, and how much each one has spent. They're not digging through settings. The information they need to stay in control is always one tap away.

User-defined policy controls

Each agent connection gets its own policy. Per-transaction cap, daily cap, monthly cap, approval threshold. Your app surfaces these as a settings screen. Grid evaluates them before any agent-initiated action executes. Below the threshold: autonomous. Above it: held for approval.

Permissions are just as granular. A user can let an agent view balances and send payments without letting it create external accounts or fund itself. When a transaction exceeds the approval threshold, the user gets a prompt with the full details and a single tap to approve or decline. Delegation doesn't transfer account ownership. The user can pause or revoke the agent at any time, instantly.

Flexible connection flows

How an agent connects depends on where it lives. Grid supports multiple flows so your users aren't forced into a single path.

CLI install. The app generates a one-liner, the user runs it in their terminal, and the Grid Wallet CLI is configured with a scoped credential tied to their policy. Best for developers already working with coding agents like Claude Code or Codex.

MCP with OAuth. Grid exposes an MCP endpoint with OAuth support. The agent initiates the handshake, the user approves in the app. No terminal required.

Agent-initiated device flow. For autonomous agents with no browser context. The agent surfaces a pairing code, the user approves it in the app, and the connection is established, similar to OAuth device authorization.

In every case, the approved scopes and spend limits are set in the app before any credential is issued.

Every action, fully traceable

Every agent-initiated action stays traceable through Grid's policy decisions, approval state, and final settlement outcome. Users see exactly what their agent did: which account it paid, what it sold, how much budget remains.

AI agents are going to move money. The only question is whether the infrastructure beneath them is built for it: real policy enforcement, real auditability, and real user control. Grid Global Accounts is.

Reach out to the Lightspark team to get access and start building.