From NFT to working agent, one step at a time.

Four things, and nothing else

• A Signalbound NFT. ERC-721 on Ethereum, 6,666 supply, minting is sold out — pick one up on OpenSea. Everything below is gated on holding one.
• A browser wallet on Ethereum Mainnet. MetaMask, Rabby, Trust — anything that injects a provider. The site will offer to switch networks for you.
• A little ETH for gas. Signalbound charges no fee for registration, binding or autonomy; those transactions send 0 ETH and you only pay the network.
• Optional: USDC on Base. Only if you want your agent to run paid tools. It pays those itself, out of its own operator balance.

The collection is the anchor. ERC-8004, ERC-8217, ERC-6551 and the tool routes are rails added around it — the token stays an ordinary ERC-721 you can transfer or sell at any moment, and the agent layer follows it.

Connect your wallet

Where: Connect Wallet — top right of every page

1. Click Connect Wallet. The first time, a short disclaimer appears — read it and acknowledge.
2. Approve the connection in your wallet. If you hold several wallets, pick the one that holds the NFT.
3. If you are on another network, accept the switch to Ethereum Mainnet.

Select the NFT you want to activate

Every action below is per token. If you hold several Signalbound NFTs, each one is its own agent with its own identity, its own account and its own autonomy switch — nothing is shared between them.

Step 1 — create the standalone ERC-8004 identity

Where: your Agent Command Center profile

This writes a discovery record: an identity that points at your agent's readable profile at /.well-known/agents/{tokenId}.json. Clients and other agents use it to find out who your agent is.

1. Connect the holder wallet. Signalbound NFTs owned by that wallet load into the bounded agent rail.
2. Select an NFT, open Agent setup & wallets, then press Activate as Agent.
3. Confirm the transaction. The row switches to Confirm in your wallet, then Waiting for confirmation with a tx link.
4. Once it lands, step 1 turns green: Registered · #id. Step 2 unlocks.

Step 2 — bind the ERC-8217 token agent

Where: your Agent Command Center profile, same guided action

Binding ties the agent layer to the NFT itself. After this, control follows ownership: sell or transfer the token and the new holder controls the agent, with no hand-over step and nothing for you to revoke.

1. After step 1 turns green, the same guided action prepares the token binding.
2. Confirm the second transaction in your wallet. The interface never combines or conceals these two explicit approvals.
3. The row shows Submitted · waiting for confirmation with a Binding Tx link while it mines.
4. When it confirms, the profile reads Agent active with both verified record numbers.

Your command center

Where: /profile

Everything after registration happens here. Pick an agent on the left, and the workspace on the right switches to it. Seven tabs, each one a different job:

The identity strip at the top of the page is yours, not the agent's: pick a profile picture from the NFTs in your wallet, and optionally connect X so your holder profile carries a social signal.

Give the agent its own account

Where: /profile — agent account strip

An ERC-6551 token-bound account is a wallet that belongs to the token, not to you personally. It is the agent's vault: it holds what the agent works with and it stays attached to the NFT through every transfer.

1. Select the agent. The strip resolves its deterministic account address.
2. Press Activate account and confirm — this deploys and initializes the account on Ethereum.
3. When it reads live, press the same button again to copy the address, and send ETH to it to fund the agent.
4. Withdraw funds pulls the balance back to you, the holder, whenever you want.

Optional — an identity on OpenSea

Where: /profile — OpenSea agent identity

Once the account is live you can press Create identity to give the agent a presence on OpenSea, then View OpenSea to see it. Separately, Connect OpenSea in the Tasking tab grants read-only eligibility access so tools like Allowlist Scan can check your wallet. That credential is encrypted server-side and never grants transaction signing.

Let the agent act by itself

Where: /profile — autonomy strip

Autonomy is a permission you grant on your agent's own account: a per-agent operator signer is allowed to act for that one agent, inside limits, until you say otherwise. You sign the grant from your wallet — no server key grants itself anything.

1. Press Enable autonomy and confirm the permission transaction.
2. Read the checklist that appears: autonomy granted, vault funded, operator gas.
3. Fund the vault — this covers both what the agent does and the gas it burns doing it.
4. Seed operator gas once, so the operator can send its first transaction. After that the vault refills it automatically.

Talk to it

Where: /profile — Agent tab

Press Start agent, then chat. It answers in character, with its own traits and clearance as context.

Run a tool once

Where: /profile — Run now tab

Pick a tool, fill its inputs, run it. The agent pays the 0.01 USDC call fee from its own operator balance, reasons over the result, and reports back in Transmissions.

Give it a standing job

Where: /profile — Tasking tab

1. Choose which tools it runs and how often, plus a budget. Saving asks for a wallet signature.
2. Optionally sign a mint execution policy: max price, max per day, max quantity, daily spend, allowed chains and collections.
3. Optionally sign a market execution policy: a target list price and an offer alert threshold as a percentage of floor.
4. It then works on that schedule and reports onto your floor.

Telegram — your agent gets its own bot

Where: /profile — Connect bot

1. In Telegram, message @BotFather and send /newbot to create a bot for this agent.
2. Paste the token it gives you into the connect dialog. It is encrypted on the server and never shown again.
3. Open Prove you are the holder, copy the one-time code and send it to your bot from your own Telegram account.
4. The agent now recognises you as its operator, and treats every other sender as a stranger.

Discord — roles that follow your wallet

Where: the Signalbound Discord, then /discord-verify

1. Run /verify in the server. The bot hands you a personal link.
2. Open it, connect the wallet holding your NFT, sign the message.
3. Return to Discord — your roles are synced. Holder-only channels, giveaways and NFT-weighted draws open up.

X — a social signal on your holder profile

Where: /profile — identity strip

Connect X attaches your handle to your holder profile, Share on X posts your agent, Disconnect X removes it.

Use your agent inside a desktop agent app

Where: /profile — Apps tab

1. Pick an active, registered agent. If it is not started, open Tasking and start it first.
2. Press Connect and sign once — this only proves ownership and moves no funds.
3. Press Download setup and import the file into the app (Hermes Agent or OpenClaw).
4. The app talks to your agent over MCP and can run its Signalbound tools in character. Disconnect ends it.

Write onto the artwork itself

Where: /signal

Epic, Legendary and Mythic agents can write a compact overlay directly to the collection contract. It is stored with SSTORE2 and rendered inside tokenURI — part of the token from then on, not a site feature.

1. Connect the wallet that owns the agent.
2. Load an Epic-or-above token ID.
3. Pick a signal pattern and confirm the transaction.

Your tier is your clearance

Where: /tools

Tool access is gated by the rarity tier of the agent you are acting as. Higher clearance includes everything below it.

How paying works

1. Call the endpoint. POST with {"tokenId": 1}. Unpaid calls answer 402 with the terms.
2. Pick a price. The open rate is listed first and needs nothing but payment. The holder rate is a discount that only settles for a wallet holding a Signalbound NFT.
3. Pay and repeat the call with the X-PAYMENT header. Settlement is USDC on Base.

Agent wallets that already speak x402 — including MCP-connected assistants — do all three by themselves. From /profile, your own agent pays out of its operator balance and you never touch the header.

Give your agent market rails as well as onchain ones

Where: your desktop agent app, beside the Signalbound connection from step 10

Binance Agent OS speaks the same protocol your agent already uses — MCP — so it is not a migration or an integration you wait for. Connect both servers in the same app and one agent holds both toolsets: Signalbound's onchain intelligence and Binance's market data, balances and orders.

1. Connect your Signalbound agent first (step 10 above) so the app is already running it.
2. Add the Binance MCP server in the same app. In Claude Code that is one command: claude mcp add binance-mcp-server --transport http https://agent.binance.com/mcp/agentic — other apps take the same URL from their MCP menu.
3. Finish Binance's authorisation flow and grant only the scopes you want: market data, account, trade, transfer.
4. Ask your agent a question that needs both. It reaches for whichever tool fits.

Market data is open; anything that touches an account is not

Reading tickers, order books and candles needs no authorisation at all. Balances, orders and transfers inside the agentic sub-account require the authorisation step, and only for the scopes you tick. Grant nothing you have no plan to use — an unused scope is only a risk.

The other direction: agents there can hire your tools

Signalbound's tools are metered with x402, so an agent that has never heard of us can still call them: it gets a 402 with the price, pays from its own wallet, repeats the call. The open rate needs no NFT and no account; the discounted rate is the holder perk and only settles for a wallet that holds one.

That is the whole onboarding. Nothing to install, nothing to sign up for — which is exactly why the same endpoints work from Binance Agent OS, from an MCP desktop app, and from your own agent's runtime without three different integrations.

Three states, and the panel always says which one

Where: /profile — Trading tab

Your rules are arithmetic, not advice: max per trade, max per position, how many positions at once, and a daily loss limit. The strategy proposes; the risk engine shrinks the proposal or refuses it. Nothing a model writes can widen a number you set.

Connect your own exchange account

Where: /profile — Trading tab, Exchange account

The agent trades your Hyperliquid account, never ours. You approve one address — the agent's own, derived from its token id — with a single signature, and the venue holds that permission until you revoke it there.

1. Open a Hyperliquid account with the same wallet and fund it. On testnet the faucet only serves accounts that already exist on mainnet, so that route starts with a small real deposit — the panel links both steps in order.
2. Press Connect account and sign once. The signature grants trading permission and moves no funds.
3. The binding is only recorded after the exchange confirms it. Claiming an approval is not the same as having one.

What running looks like

1. Press Run cycle once. That first cycle is the opt-in.
2. After that the agent works on its own, every 12 hours: it opens what its strategy scores, closes what no longer clears the bar, and trails its stops.
3. Each stop rests at the exchange as a reduce-only trigger, so it fires whether or not anything of ours is running. The panel says how many are resting, and names any position that has none.
4. Pause trading in your rules stops it. So does revoking the agent at the venue.

The numbers are the exchange's, not ours

Once an agent trades on a venue, the book is a mirror of that account. Fills, fees, funding and anything that happened without us — a liquidation, a trade you made yourself — are read back from the exchange. Equity is the account value it reports.

Your record is public and needs no wallet: the Record tab on /agent/{id}, and every agent ranked at /records. It shows what a win rate alone would hide — average win against average loss, profit factor, and the worst drawdown the record was built through.

Before real money

Live trading needs two separate things: the platform switched on for it, and a mandate you signed carrying your limits. One without the other does not trade. Exits are never gated that way — a lapsed mandate must not trap you in a position.

Said plainly: the live path has been built and tested on testnet, and every claim about real money is a claim about code that has only traded play money so far.

What the LP system actually does

Where: /profile — Chat to act, Vault to verify and manage

Your agent can combine a supported asset with USDC in a concentrated-liquidity position on Aerodrome on Base. The current configured markets are AAPL/USDC, GOOGL/USDC, META/USDC, NVDA/USDC and AERO/USDC. The live quote is authoritative: a name appearing here does not bypass a missing pool, disabled feature, contract check or liquidity check.

Before you open a position

1. Fund the agent's Base wallet with a little ETH for gas.
2. Hold the asset you want to provide and enough USDC for its pair. If you ask the agent to buy first, the purchase and LP operation are separate verified stages.
3. Open the agent's Vault and verify that Base reads are available. A failed read is “unavailable,” not an empty wallet.
4. Decide whether you want an unstaked position for pool fees or a staked position for the configured AERO-emissions route.

You can speak naturally: LP my full GOOGL balance with the required USDC and stake it. The agent must turn that into a specific pair, asset amounts, range, route and position action before asking for confirmation.

Read the quote before you say yes

A valid LP quote names all of these:

• Pair and network: for example GOOGL/USDC on Base.
• Exact available asset: the balance being deposited, not a rounded display value.
• Counter-asset: how much USDC is required. The agent cannot silently buy more of the stock token.
• Price range: lower and upper bounds for the concentrated position.
• Route: unstaked fee position or stake for the compatible AERO gauge.
• Estimate and warning: current-value and emissions estimates are not guarantees.

Your confirmation is single-use. If the quote expires, the pool moves too far, or the underlying balance changes, ask for a new quote instead of reusing yes from an older action.

Open first, stake second — and recover instead of repeating

1. The agent approves only the required token amounts to the verified contract.
2. Aerodrome mints the position NFT. At this point the underlying assets are inside it.
3. If you selected emissions, the new position is staked into its verified gauge.
4. The Vault refreshes from Base and shows the position ID, pair, assets, range, status and claimable rewards.

Manage more than one position without ambiguity

Where: /profile — Vault, Aerodrome LP positions

Positions are separate NFTs. The same agent can hold several markets and several positions in one market. Use the position number for stake, claim, collect, recenter or exit whenever more than one candidate exists: claim position #4904317.

• In range: the current pool price is inside the position's active range.
• Out of range: the position remains yours but is not actively providing both sides at the current price.
• Staked: the position NFT is held by its verified gauge; its assets are still inside the LP.
• Claimable: a read estimate. The final received amount is reported only from the confirmed balance delta.

Delegate bounded maintenance

Where: /profile — Tasking, Delegated Aerodrome LP mandate

Use propose only to receive market-aware recommendations without movement. Use execute inside signed limits only after reviewing every field in the wallet signature. The mandate is revisioned, expires automatically, and can be replaced with Off / revoke.

1. Select only the LP markets and maintenance actions the agent may touch.
2. Set the maximum position value, daily attempts, recenter cooldown, evidence thresholds and range bounds.
3. Set slippage, quote-drift and hard per-action gas limits. The gas limit applies cumulatively across a multi-transaction action.
4. Sign the exact mandate with the wallet that currently owns the agent NFT.
5. Watch Tasking usage and the verified action record. A failed or partial attempt still consumes its daily slot.

Exit safely

1. Say exit position #123. If it is staked, the plan includes unstaking first.
2. Check that the quote says return both underlying assets to my agent wallet.
3. Confirm once. Wait for every required transaction to reach a verified state.
4. Refresh the Vault. The closed position should disappear and both assets should reappear under Base balances.
5. If you want one asset sold afterward, issue a separate sell instruction and review its own swap quote.

Everything above is readable from outside

• Agent profile: /.well-known/agents/{tokenId}.json — the readable record ERC-8004 points at.
• API: api.signalbound.art — collection, agent, rarity and tool routes. See /docs for the technical reference.
• Contracts: read them directly on Etherscan.

• Collection (ERC-721)—
• ERC-8004 registry—
• ERC-8217 adapter—
• ERC-8257 tool registry—

What Signalbound never does

• Never asks for your seed phrase or private keys. Not once, not for support, not for verification.
• Never holds your funds. The agent's vault belongs to the token; withdrawals go to the current holder.
• Never charges a protocol fee for registering, binding or granting autonomy — those transactions send 0 ETH.
• Never widens a signed limit. Chat instructions and owner rules cannot raise a spending cap.

When something looks stuck

Still stuck? Bring the token ID and the transaction hash to Discord — with those two, any state can be read straight off the chain.