Console: run an agent in plain language

The console is the plain-language on-ramp to the H-Series. You describe what you need; it finds a matching service in the H-Index registry (or connects to any public MCP server you name), shows you what that service can do, and recommends a governed agent to use it. You create the agent, give it a budget (an owner-signed on-chain H-Grant cap), fund it, and run tools under it. Every paid run is authorized, capped, and receipted.

What it is

The console at h-series.xr-utilities.com/console has three parts on one page:

• The chat. Describe a task or a service in plain language. The assistant searches H-Index for a fit, or you paste an MCP server URL to connect directly. It presents the service’s tools as cards and, when you pick one, recommends an agent template. The chat itself never creates an agent and never spends.
• Your agents. Where you actually create an agent, set its budget, fund it, and run tools. This is the only surface that acts.
• Rings. Membership rings (H-Pact) you belong to or manage.

Two owner models sit behind the agents:

• Google sign-in (embedded). H-Agent mints and manages the agent’s account for you. No wallet needed. You fund it by sending USDC on Base to the account address, and you can withdraw the unspent balance later.
• Wallet (connected). Your own connected wallet is the funds owner and grants the agent its cap directly.

You can hold both at once; they are distinct owners, so you pick which one owns a given agent.

How to use it

1. Open the console at h-series.xr-utilities.com/console. Browsing is open; you sign in when you are ready to create an agent.
2. Describe what you need in the chat, or start from a template, or paste an MCP server URL in the Connect bar. The assistant finds a service (searching H-Index) or connects the server, and lists what it can do. Free and paid tools are labeled with their price.
3. Sign in with Google or connect a wallet from the top bar. Either works on its own.
4. Create an agent in the “Your agents” panel. The chat’s recommendation prefills the template; you can also create a custom agent. With Google, H-Agent runs the account; with a wallet, your wallet funds it.
5. Set a budget. Enter a daily USD amount and the console publishes and applies an H-Grant grant naming the agent as grantee with that cap. Free tools run without a budget; a budget unlocks paid tools. (Advanced: apply an existing grant id, or use a wallet owner’s grant.)
6. Fund it (Google/embedded owners). Send USDC on Base (network eip155:8453) to the agent’s account address shown in the Fund box. A budget is a ceiling, not a balance: a paid run draws real USDC from this account. A wallet owner funds from their own wallet.
7. Run a tool. Free tools run immediately on a signed-in agent. A paid run shows the price and asks you to confirm (“Run for ~$X? This spends real USDC…”); on confirm, the agent pays over x402, capped by its on-chain budget, and the result plus a receipt reference come back.

For agents / paid runs

The chat is a starter, not an executor. It never spends. All execution happens on the agent, and a paid run has three preconditions, enforced independently:

• Signed in (Google or wallet), the identity the agent is owned by.
• A budget (an owner-signed on-chain cap, via H-Grant). This is the ceiling.
• Funding (real USDC on Base in the agent’s account). A budgeted-but-unfunded agent shows “ready” but fails a paid run at settlement, because a budget authorizes spend, it is not a balance.

The first paid run of a session prompts a confirm dialog with the quoted price before anything settles. The per-run price is also bounded by the agent’s on-chain cap, so a settlement can never exceed the remaining budget for the period. If a service’s own payment check refuses the signature, or the price is above the per-run limit, or the remaining cap is too low, the run is refused and nothing is spent; the console explains which of these happened.

Under the hood, a run posts to H-Agent (h-agent.xr-utilities.ai), which pays the service over x402 and anchors a sponsored H-Seal receipt. The agent connection token can also be used from your own MCP client outside the console.

Good to know

• Chats are device-local by default. They live in your browser’s local storage, not on a server. Turn on cross-device sync (signed in) to mirror them to your account with PII redacted before storage. No keys or sign-in tokens are ever stored in the browser.
• Run history is recorded to your account when signed in, so it shows on every device you sign in on. A run’s full saved output stays on the device that ran it; the on-chain receipt is the durable record.
• Funding is Base USDC only. Withdrawing the unspent balance back out is available for Google-managed (embedded) accounts and requires a fresh sign-in (step-up) to move funds.
• Some templates are previews. A template marked “preview” names a capability built in the estate but not yet runnable from the console; it does not start a run.
• The assistant needs a model key configured on the deployment; if it is not, the chat says so and connecting/browsing still works.

Where to go next

• New to the suite? Read Start here for the lifecycle and how the products compose.
• The budget you set is an H-Grant cap. See how grants, custody, and revocation work in the H-Grant guide.
• The agent identity and its spend cap are H-Agent. See the H-Agent guide.
• The services you find come from the registry. See the H-Index guide.