This page sets out the risks. It is not investment advice, and Alokumi does not promise any return.
Market risk
- Concentrated liquidity can lose money. When prices move, a position incurs impermanent loss: the ratio of the two tokens you get back on exit differs from what you put in, and the value can be lower than “just holding”. Fees do not necessarily make up for it.
- Out of range means no fees. Once the price leaves the range, the position is entirely one token: if the price is falling you are holding the token that is falling; if it is rising, everything has been sold into the quote token.
- Meme tokens can go to roughly zero quickly. Many high-fee pools hold new or small-cap tokens whose price can lose most of its value, or nearly all of it, within minutes, and liquidity can disappear at the same time.
A stop-loss does not guarantee the fill price
- Stop-loss, price stop, and max hold are checked roughly every minute, and the exit transaction is sent after a trigger. Between two checks, and before the transaction lands on chain, the price keeps moving.
- In a fast crash the fill can be much worse than the stop line. In our own live-funds data, with a 10% stop-loss line the median stop-loss exit was about −12%, and some were worse than −15%.
- When a valuation is temporarily unavailable, the stop-loss waits; it does not exit just because data is missing.
- Tokens left after an exit are sold back into the quote token according to Swap back leftovers after exit, and that swap has slippage too.
The AI can misjudge
Each round the agent looks up data and makes judgments from your prompt. It may pick the wrong pool, enter at a bad moment, place an unsuitable range, or misread your words. The safeguards:
- The agent can only pick from pools actually found this round, and its queries can only be stricter than the conditions you fixed;
- Every open must pass safety checks, a sell stress test, and an authorization check; if any fails, nothing is opened;
- Protective exits — stop-loss, price stop, max hold, a token becoming unsafe, a close you requested — are executed by a deterministic engine, and the AI has no veto;
- When the AI is unavailable, times out, or returns invalid output, no position is opened that round, existing positions are treated as “keep holding”, and protective exits run as usual.
These rules limit what the AI can do; they do not guarantee that it does it well. See How it decides and Security.
The Vault and the platform
- The Vault program has not been externally audited. It is deployed on Solana mainnet as a test version, and its upgrade authority sits in a Squads multisig. A problem in the program code or with the upgrade authority could cause loss of funds. Only deposit an amount you can afford to lose entirely.
- The platform and its infrastructure can fail. When external services such as Solana RPC, the model gateway, or Jupiter quotes and prices are unavailable, the agent skips that round or step rather than executing at risk; during that time positions may not be managed promptly. A real-funds agent that has not completed a check round for more than 15 minutes sends an “Agent stopped checking positions” alert.
- Your exit does not depend on the platform. The Emergency exit (no platform needed) actions in the Funds tab are signed by your wallet one by one, so you can claim fees, close positions and withdraw yourself even if the platform is down. See Use real funds · Emergency exit.
- The Vault depends on two external programs, Meteora DLMM and Jupiter; if they change their interfaces or are disabled, closing positions or swapping may fail.
Paper funds and backtests do not predict real returns
Paper funds use live market data, but fees are estimated from pool data and fills are booked as estimates, without the failures, delays, and fill deviations of real transactions. The same goes for dry runs and the 14-day backtest (and the 30 / 90-day ones): backtest fees depend on liquidity snapshots, and when those are incomplete the estimate can be far off; the app marks it. The official paper Agents’ public results are paper funds too. Paper results, dry runs, backtests and any past performance do not represent real-funds returns.
Copying smart money doesn’t mean you’ll earn the same
The leaderboard is past data. Their position may close before you see it, and what happens after you copy is your own result: your price, timing and amount all differ from theirs.
Fees at a glance
| Item | Fee |
|---|---|
| Viewing positions, the smart-money leaderboard, paper funds | Free |
| Swaps we route when you open / exit from your wallet | 0.2% of the swapped amount |
| Agent with real funds | A share of claimed LP fees, tiered by pool, never the principal in your positions; charged even when the position as a whole lost money; depositing to and withdrawing from the Vault also cost 0.1% of the amount |
| On-chain rent and network fees | About 0.0015 SOL once to register your account (not refundable); about 0.004 SOL to create a Vault (not refundable for now); about 0.04 SOL per position (refunded on close); about 0.07 SOL per new bin array (not refunded) |
The fee table (read live from the on-chain Vault config) and details of each item: Fees.
Compliance
The platform has not yet resolved matters such as custody licensing, KYC/AML, and sanctions screening. You are responsible for confirming whether using this kind of service is legal where you live.
Suggested way to use it
- Run on paper funds for a while first and see whether the pools the agent picks, how often it resets ranges, and its exit timing match your playbook.
- When switching to real funds, start small and only commit an amount you can afford to lose entirely.
- Keep some SOL in the Vault for rent, and connect Telegram to receive exit and issue alerts.
- Know where Emergency exit (no platform needed) is, and how to pause the agent and the Vault.