Skip to content
Documentation
What every part of Agentic Hub does, how to use it, and where it stops being reliable. Written against the running build, not a plan.
Agentic Hub · BNB Chain · last updated 25 August 2026
What this is
Agentic Hub is a marketplace of agents that read BNB Chain on demand. You give an agent an address, it reads public chain data and public market APIs, and it answers in seconds. There is no account, nothing to install, and no subscription. You pay per run.
Two other things live here. A board where anyone can post funded work, and a listing form so people who have built their own agent can put it in the same catalogue.
Running an agent
- Pick an agent from the catalogue. Each card shows what it answers, what input it needs and what a run costs.
- Enter the input. Most agents want a BNB Chain token address. The field validates the shape before it lets you run, so a typo fails immediately rather than after you have paid.
- Connect your wallet. The site asks for your address and for BNB Chain, and nothing else.
- Confirm the transfer. Your wallet shows the amount and the destination. Nothing happens until you approve it in the wallet.
- Read the answer. The run starts once the transaction is confirmed on chain.
What the wallet connection can and cannot do
The connection is read-only. The site reads the address you choose to share and asks your wallet to switch to BNB Chain (chain id 56) if it is on another network. It cannot move funds, and it never asks for a private key or a seed phrase. No legitimate site ever will, including this one. You can disconnect at any time from the wallet control in the header, which clears the address on our side; revoking the site in your wallet's own connection list removes it there too.
Where the money goes
The payment is an ordinary BNB transfer from your wallet straight to the project address. It does not pass through the site, and the site cannot hold, forward or refund it. Once your transaction is confirmed it is on a public blockchain and nobody can reverse it.
Each payment can be used for exactly one run. A transaction hash that has already paid for a run is rejected if it is presented again.
The agents
Ten agents run here, built and maintained by us. Prices are per run.
| Agent | What it answers | Input | Price |
|---|---|---|---|
| Contract check | Traps in the token contract, snipers at open, what happened to the liquidity | token address | $1.00 |
| Wallet clusters | Bundle signatures: throwaway wallets, synchronised buys, identical entry sizes | token address | $1.00 |
| Who holds it | The wallets big enough to move the price, and what they hold now | token address | $0.75 |
| Liquidity moves | Liquidity added and pulled, including a slow drain before it shows on the chart | token address | $0.75 |
| Price impact | How far a sell of a given size moves the price, against real depth | token address and sell size in USD | $0.50 |
| Ticker twins | Other tokens sharing the ticker, and which one is the real one | token address | $0.50 |
| Compare two tokens | Two tokens side by side on money, depth, activity and contract risk | two token addresses | $0.50 |
| Wallet activity | What an address has been doing lately, in and out, across tokens | wallet address | $0.50 |
| Fresh launches | Pairs created on PancakeSwap in the last minutes | how many recent blocks to scan | $0.35 |
| Transaction breakdown | What a transaction actually did, in plain words | transaction hash | $0.25 |
How prices are charged
Prices are set in dollars and converted to BNB at the rate at the moment you run. The amount your wallet asks you to confirm carries about two percent of headroom, so that a small move in the rate between the quote and your confirmation does not cause the payment to be rejected. The check on our side accepts anything at or above ninety-seven percent of the quoted amount, so that headroom is not a fee and you are not charged twice for the same volatility.
What an answer is worth
Every agent computes from public data at the moment you ask. That has hard limits, and it is better to know them before you rely on an answer.
- It is not an audit and not advice. A clean result means the specific checks did not fire. It does not mean a token is safe or that the people behind it are honest.
- History is shallow. Free public nodes serve only part of the archive, and none of them serve historical account state. For tokens older than roughly a year, some questions cannot be answered at all from free infrastructure.
- Sources go down. Market data comes from DexScreener, and falls back to GeckoTerminal when DexScreener is rate limiting. When neither responds, the agent says so instead of guessing.
- An agent that cannot establish something says so. Absence of a warning is not the same as a clean bill of health.
Posting a task
The board is for work you want done that no agent here covers. You describe the work, set a reward and a deadline, and leave a contact so whoever wants the job can reach you.
Five award modes describe how the reward is decided. The mode is a description of your intent, not something the site enforces.
- Bounty — many people attempt the work and one result is paid.
- Claim — one worker reserves the task before starting.
- Pitch — workers propose a plan before any delivery work begins.
- Benchmark — a measured result decides who gets paid.
- Auction — workers compete on price, timing or allocation.
Be clear about two things. There is no escrow: the site does not hold the reward and does not release it. And posting creates no obligation on anybody. Whatever you agree with the person who answers is between the two of you, and the site is not a party to it and does not arbitrate it. Your contact is stored privately and is not shown on the public board.
Listing your own agent
If you have built an agent, it can appear in the same catalogue, at a price you set. Your agent runs on your own infrastructure; we call it and show what it returns.
The endpoint contract
You give us one URL. We send it a POST with JSON:
POST https://your-endpoint.example/run
Content-Type: application/json
{ "input": "0x…", "extra": "optional second field" }
Answer with HTTP 200 and JSON in this shape:
{
"headline": "one sentence a person can read",
"lines": ["supporting detail", "supporting detail"]
}
Rules that are actually enforced:
- Twenty five seconds. Past that the run is reported to the visitor as your endpoint timing out.
- Status 200. Any other status is shown as your agent failing.
- Your URL stays private. We call it server-side; visitors never see it.
- Public hosts only. Endpoints pointing at private or loopback addresses are rejected, so the site cannot be used to reach into anyone's internal network.
- Price between $0.05 and $25.00 per run.
There is a working example you can point a listing at before you write anything: /api/demo-agent. Send it a GET and it returns this contract; send it a POST with a BNB Chain token address and it returns real market data in the expected shape.
Review, and how you get paid
Listings are reviewed by hand before they go live, because the endpoint is called with our name on it. Submit the listing with a contact we can reach; if something is wrong with the endpoint we will tell you rather than silently dropping it. Be honest that this is a small operation: review is a person reading your submission, not an automated pipeline, and it is not instant.
Payouts to operators are arranged directly and are not automated yet. Do not list an agent expecting an automatic revenue split today.
How payment is verified
When you submit a run with a payment, the transaction is checked before any work is done. It has to pass all of these:
- the hash is well formed, and has not already been used for another run;
- the transaction exists on BNB Chain and has a receipt;
- the receipt status is success, not a reverted transaction;
- the recipient is the project address;
- the value is at least the minimum for that agent's price.
If any check fails, the run does not start and you are told which one failed.
Limits worth knowing
- Everything here runs on free public infrastructure. During periods when the public nodes are degraded, some agents are slower or return partial results.
- A single run has a hard ceiling of sixty seconds. Agents that scan block ranges narrow their scan to fit rather than returning nothing.
- Tasks on the board are not routed to agents automatically. Operators read the board and get in touch with posters themselves.
Questions
Reach us on X at @tryagentichub. If something in this document does not match what the site actually does, the document is the thing that is wrong, and we want to know.