Mycelium: an on-chain-only trading agent for the Mnodes network

Mycelium is an experimental, non-custodial trading agent for Robinhood Chain. It reads public blockchain data, turns it into ranked, actionable trade ideas, each with an entry, a stop, a target and a verified trading route, and hands each idea to a member who decides whether to sign it. Ideas that cannot actually be traded never reach the board. Nothing else is used as input: no social media, no news, no sentiment, no paid listings.

Access is granted by a single asset: an Mnode NFT from the Mnodes collection. The NFT is the membership card, the pseudonymous identity and the unit of standing inside the network. There is no email, no password and no subscription.

The second idea behind the project is collective: every finished trade a member makes is written into one shared journal, and that journal calibrates how much the tool trusts each strategy. A new member therefore starts with everything the network has already learned, instead of from zero.

This is an experiment, not a financial product. It gives no financial advice. Every member carries one hundred percent of the risk and the outcome of every trade they choose to take.

Most retail trading decisions are made from crowd mood: a post, a screenshot, a headline, an influencer call. Crowd mood is cheap to manufacture, arrives after the move, and is the main reason beginners lose money.

Meanwhile the information that actually matters is already public and free. On a chain, the pool depth, every swap, every holder, every contract deployment and every whale movement is verifiable. It is simply too fast and too fragmented for a person to read by hand.

The third problem is memory. Individual traders repeat mistakes because their own sample is tiny. A group of traders using the same tool has a far larger sample, but normally no way to pool it without also pooling their hype.

Mycelium takes three positions and holds them without exception.

• On-chain only. If a claim cannot be proven by a transaction, a pool balance or a contract, it never reaches a member's board.
• Explained or nothing. Every idea shows the evidence, the conviction, the concerns, the stop, the target and what would prove it wrong.
• The member signs. The tool proposes, the wallet decides. Funds never leave the member's control.

The system is a continuously running read of Robinhood Chain (chain id 4663) feeding a deterministic strategy engine, wrapped in a web app that never holds keys or funds.

• Ingestion. New blocks are read continuously. Swaps, transfers, contract creations and pool state are decoded and stored as metrics per token: price, volume, trade count, pool depth, holders, holder change, buy versus sell pressure.
• Token intelligence. Each token is checked against contract facts: verified source, mint ability, transfer or sell blocking, deployer holdings, supply concentration, age.
• Strategy engine. A set of rule-based playbooks scores every token independently. The engine is deterministic and unit tested, so the same data always produces the same idea. No numbers are invented by a language model.
• Quality gate. Ideas are scored after estimated fees and slippage, then filtered on expectancy, break-even win rate, a minimum reward to risk of 1.8, liquidity floors and the measured record of that playbook. Only ideas at or above 80 conviction reach the central Trade Ideas board.
• Risk engine. Position size is edge proportional, capped per trade and in total, with a cooldown and a loss-streak halt.
• Execution. In manual mode the member's wallet signs each swap. In autopilot a local burner wallet, whose key stays on the member's device, signs inside the limits the member set.
• Watchtower. A backend job detects stop and target conditions even when the browser is closed. It cannot sign a sale. The exit swap runs after the app reopens on the device holding the wallet key.
• Assistant. An in-app trading assistant answers questions about the tool and about trading, grounded in tool calls against live chain data, and always ends with the reminder that this is not advice.

Several families of setup are built from chain data, each with its own evidence and its own failure mode. They score independently and all of them pass through the same quality gate before anything reaches the central board.

• Momentum and breakouts. Price clears a level on real volume with a widening set of buyers. Fails on false breaks, so it needs a tight stop.
• Following informed wallets. Wallets with a measurable record are seen accumulating in real time. Fails when they are hedged, early by weeks, or exiting into you.
• Fresh launches. Contract creation, first liquidity and first buyers are visible within seconds. Genuine early access, and also where nearly all rug pulls live, so hard safety checks apply and size stays tiny.
• Value dips on solid coins. A coin with deep liquidity, real volume, a broad holder base and a track record that has fallen hard can snap back. Solidity is measured from chain data, and the fall must show signs of stabilising before the idea qualifies. Recovery timing is given as a rough band, never a promise.
• Pool watch. Opening commitment, depth trend, turnover and pull risk read straight from pool state. Draining, high-risk and thin pools are rejected outright.
• Flush reversion. A sharp intraday flush that still shows relative volume, stabilisation, buy pressure and a real liquidity floor can revert. Everything thinner than that is discarded.
• Chain strength. Breadth, median move and tape regime across the whole chain, used as its own read and as a filter on every other idea, because a good setup in a bad tape is still a bad trade.

Entering is the easy half. The same rules run in the browser and in a backend watchtower. The watchtower can flag an exit while the laptop is closed, but the actual sale waits for the app and its locally held signing key.

Size itself is derived rather than guessed: bankroll, a member-set risk percentage per trade defaulting to one percent, the stop distance, conviction and pool depth produce a suggested amount. It stays a suggestion, and the member sets the final amount, stop and target on every single idea.

A chain pool has no native stop order. Every exit is a swap that has to be signed, so three honest safeguards sit around the level: a small safety margin that fires slightly before the stop, an arm switch that lets an open position ask for the exit signature by itself while the app is open, and a warning as price approaches the stop or the target. In a violent gap or a thin pool the fill can still be worse than the level, and that is stated rather than hidden.

• Break-even shift once a trade has earned it, then a trailing stop.
• Partial profit taking, with the peak and the scaled-out portion remembered.
• A time stop, because a trade that goes nowhere is still costing exposure.
• A stop safety margin and an approach warning distance, both member adjustable under wallet protection.
• Arming is per position and optional. An unarmed position only warns; nothing is signed without the member.
• Any position can be closed at any moment: sell back into USDG, keep the coin and close the record, or force-close the record if the sale fails on chain.

Trades are executed directly against the chain's own liquidity, quoting through Uniswap V3 QuoterV2 and submitting through SwapRouter02. There is no aggregator in the path, no third-party sign-in and no relayer holding an order on a member's behalf.

When direct liquidity is poor, the tool builds its own route: it compares every direct fee tier against two-hop paths through hub tokens derived from recent chain activity, quotes the whole path and executes the better one. The interface says when a conversion is routed through a middle token. Slippage bounds and minimum received are enforced inside the transaction, and the signature always comes from the member's own wallet.

The Mycelium takes its name from the living network under a real forest: separate fungi connected underground, moving nutrients and warnings between trees. Inside the app that shared brain of all members is called the CMN, the Common Mycorrhizal Network, and it works in three strictly separated layers.

• Layer 1, ideas. Every agent idea comes from chain data alone, never from what other members did. A network that copies itself mistakes luck for edge.
• Layer 2, discipline. Finished member trades calibrate which playbooks are trusted and what a setup really needs after costs. Each member counts for at most ten recent trades per asset, so no wallet can bend what the network learns.
• How that evidence is weighed. A finished trade decays as it ages, carrying half its weight after roughly three weeks, so recent results outrank last quarter's. A trade with an on-chain transaction hash counts in full, while one that cannot be checked against the chain counts for about a third. Every trade is also stamped with the kind of market it closed in, rising, mixed or falling, and a playbook is trusted according to how it behaved in the market the network is in right now.
• Layer 3, inspiration. A member's winning trade can be opened as an editable draft at today's price, clearly marked as a member play, never signed automatically.
• Standing. A score from 0 to 100 built only from finished trades: win rate, average result, breadth, recency and how much the member fed the shared memory. Trades carrying an on-chain transaction hash weigh more. Five finished trades are required before a member appears.
• Pulse, seasons and pods. Anonymised network activity, fresh monthly seasons with streaks, and small private circles of up to twelve members who compare results under their Mnode numbers.

Membership is the NFT. A wallet holding an Mnode NFT from the Mnodes collection connects, signs one free message, and enters. The check is repeated on every visit, so if the NFT leaves the wallet, access goes with it.

Everywhere the network is shown, members appear as their token number, for example Mnode #412. A token number is public on chain and says nothing about a person, so the network stays pseudonymous while keeping a stable, unforgeable identity. Wallet addresses, balances and settings are never shown to other members.

There is no subscription and no fee taken on trades. Holding the NFT is the only condition of use.

Mycelium is non-custodial. It sees a public address and proposes transactions. It cannot move funds, and it never receives a private key of a connected wallet.

The always-on wallet is the single, opt-in exception. A member can create a second wallet whose private key is generated on the server, encrypted with AES-GCM under a project secret and stored in a service-role-only table. It exists so a position can be closed at its stop or target while the browser is shut. It only closes trades by default, it is bound by the member's own per-trade, open-risk and daily-loss limits, withdrawals are restricted to the wallet that created it, every signed action is logged with its transaction hash, and the member can export the key or delete the wallet at any time.

Autopilot uses a separate burner wallet created in the member's browser. That key lives on the member's device, so an autopilot wallet should only ever be funded with an amount the member is comfortable leaving to a browser-held key. Autopilot can be halted at any second.

Personal data is minimal by design: positions, balances and settings live in the member's own browser, with optional encrypted cloud backup. Only finished trade outcomes, never addresses, feed the shared journal.

No strategy wins every trade, and no reading of chain data predicts the future. What chain data does well is describe the present precisely: how deep a pool is, who is buying, whether a contract can rug you. What it cannot do is tell you what happens next.

The known limits are stated plainly: thin markets can be manipulated, young tokens can lose most of their value in minutes, an empty board is a legitimate result rather than a bug, and past performance of a playbook is a record, not a forecast.

Layered protections always apply: per-trade cap, total exposure cap, a stop and a target on every position, a loss-streak breaker, a cooldown against revenge trading, and hard filters against thin liquidity, extreme slippage and unverified contracts.

On top of those, each member can set their own wallet protection: a rolling 24 hour spend budget and a rolling 24 hour realised loss limit. Manual trading may override a single block after an explicit warning, autopilot may not, and closing a position is never blocked by a limit.

Mycelium is experimental software provided as is. It is not a broker, not an exchange, not a fund and not an adviser. It does not manage money and it does not hold money.

Nothing produced by the tool, the assistant, the network pages or this document is financial, investment, tax or legal advice. NFA. DYOR.

Every member is one hundred percent responsible for every trade they accept and for every amount they commit. Losses can be total. Only funds a member can afford to lose should ever be used.

Holding an Mnode NFT grants access to an experimental tool. It is not a share, not a security, not a promise of profit, revenue, dividends, buybacks or future value, and no return of any kind is offered or implied.

Members are responsible for complying with the laws and tax rules that apply where they live. The full wording is on the Terms and risk page.

The direction of travel is more evidence and more shared memory, not more noise. Deeper contract screening, richer recovery-timing models built from historical chain behaviour, more strategy surfaces that all feed the same curated Trade Ideas board through the same quality gate, and stronger proof-of-trade verification so standing is fully provable on chain.

Anything that would introduce social signals, hype feeds or copy-trading automation is deliberately out of scope, permanently.

The project stores as little about a person as a members-only tool can. There is no account, no name, no email address and no password. Identity is a wallet address holding an Mnode NFT, plus the settings and trade records attached to it.

A wallet address can be personal data under the EU General Data Protection Regulation, so it is treated that way: row level security keeps each member's records reachable only by that member, keys never leave the member's device, and nothing is sold or shared for marketing. Finished trades are published to the network under an Mnode number rather than an address, while acknowledging openly that a public ledger makes pseudonymity, not anonymity.

Members can request access, correction, portability, deletion, restriction, objection and withdrawal of consent. Where a finished trade already calibrated the shared record it is anonymised rather than removed, so the network's statistics stay honest. The full policy is on the privacy page.

The metaphor is structural, not decorative. A chain is soil: continuously active, public, and mostly unread. A member is an Mnode: it responds only to what it can physically reach. The Common Mycorrhizal Network is the web between them, carrying what a finished trade taught to every other node, exactly as a forest passes nutrients and warnings underground.

The design follows from it. Ideas light up only when a signal reaches them, losses are composted into shared learning instead of hidden, and there is no central tree deciding anything, because the colony keeps its own keys.