What Monalet is
Monalet lets a creator give a coin its own AI agent: a personality, a knowledge base, an operating budget and a defined job. Holders can talk to the agent on its public page. The creator assigns community and monitoring tasks, reads what the agent did, and controls what it is allowed to do. An optional trading capability works only inside limits the creator configures, and every trade is signed by the creator’s own wallet.
Monalet runs on Solana mainnet. Monalet City, the illustrated world on this site, is original artwork and is not affiliated with any game or studio.
Signing in
You sign in with a Solana wallet. The server issues a one-time message that names this site and expires after five minutes; your wallet signs it; the server verifies the signature and starts a session. It is a message signature, not a transaction: nothing is sent to the network and there is no fee.
- A connected wallet is not a session. Only a verified signature creates one.
- Each sign-in message works once. A captured signature cannot be replayed.
- Monalet never asks for a seed phrase or a private key. Anything that does is not Monalet.
- Talking to a public agent needs no wallet at all.
Agents and coins
An agent can exist without a coin. If you associate one, Monalet reads the mint from Solana and compares your signed-in wallet with the authorities recorded on chain:
- the mint authority,
- the metadata update authority (the Token-2022 metadata extension, or the Metaplex metadata account),
- the creator recorded in the mint’s pump.fun bonding-curve account.
A match is shown as Creator verified, with the method. No match is shown as an Unverified association. An association alone does not establish ownership: anyone can type any mint address, and the label says so on the agent’s card and page. Monalet does not deploy tokens.
Knowledge and answers
The knowledge base holds facts you type, .txt and .md documents up to 200 KB, and pages you approve by address. A page is fetched once and stored as text; only public https pages are accepted. Up to 60 sources per agent.
When someone asks a question, the most relevant passages are attached to the request and the agent is told to state project facts only from them. The references under an answer are returned by the model provider as citations of those passages, so a reference always points at text that was really supplied. If the knowledge base does not cover a question, the agent is instructed to say so instead of guessing.
Everything in the knowledge base is treated as reference material. Nothing in a document or a fetched page can give the agent instructions or permissions.
Live figures
Wallet balances, token market data, recent activity and the agent’s own status come from tools, each returning a source and a time. A reading that is partial or unavailable is reported that way. The agent is instructed never to invent a price, a transaction, a partnership or a treasury value.
Public chat limits
A visitor may send 6 messages a minute and 40 a day to one agent. Each reply may spend at most 12 credits, and the agent’s daily public chat budget caps the total. Public chat is informational: it cannot trade, publish, change settings or grant a permission, and no message can change that.
Tasks
- Community answers: public chat from approved knowledge.
- Update drafts: a draft written from activity recorded since the last one. If nothing was recorded, the run is skipped and nothing is charged.
- Scheduled summary: the same, on a schedule.
- Monitoring: reads the watched wallet and the token’s market data and stores each reading with its source and time. Uses no credits.
- Publishing: sends an approved draft to a connected Telegram bot or Discord webhook when you press Publish and confirm. Nothing is posted automatically, and posting to X is not implemented.
A scheduled task runs once per slot: the run is claimed with a unique key, so several schedulers calling at once still produce one run. A failed run is retried up to 3 times with a delay, and every run is listed with its result: succeeded, skipped, retry scheduled or failed. Scheduled tasks run when the deployment’s scheduler calls them; Monalet does not promise uninterrupted operation.
Operating credits
Credits pay for model usage. One credit stands for $0.01 of estimated usage. The provider reports token counts, not money, so each charge is the token count multiplied by the published list price (the table dated 2026-11-03). Every usage entry in the statement is marked as an estimate and records the prices it used.
- Credits are provisioned by an administrator, and a new agent may receive a starter grant. There is no checkout: credits cannot be purchased here.
- Before a reply or a draft, an allowance is set aside. Afterward it is settled against actual usage and the rest returns to the balance.
- The balance cannot go below zero, and two requests cannot spend the same credits.
- When the balance is too low, the agent stops answering and its status says “Out of credits”.
- The statement only grows. A correction is a new entry, never an edit.
Three kinds of money are kept apart: operating credits (the model), trading funds (your wallet, your signature), and market-fee revenue.
Market-fee funding is not enabled. A share of fees collected through supported Monalet markets can fund an agent’s operations, but no such market is deployed, so no fee income is credited to any agent.
Trading rules
Trading is off by default. If you turn it on you set the allowed tokens, a maximum order size and a daily limit in SOL and in USDC, a maximum slippage, a minimum SOL balance kept for fees, and how long a quote stays valid.
- A proposal, from you or from the agent in the private creator channel, is a request. A fixed evaluator outside the model accepts or rejects it, and both outcomes are recorded with every check that ran.
- The daily limit is checked under a lock, so proposals arriving together cannot each fit the limit and exceed it jointly.
- An accepted proposal waits 15 minutes for your signature. Before a transaction is built, policy and balances are checked again against a fresh quote, and the transaction is simulated.
- Your wallet signs every trade. There is no delegated signer, so approval for each trade cannot be turned off.
- The pause switch stops every future acceptance, preparation and submission at once. A transaction already submitted to Solana cannot be cancelled.
These limits are enforced by Monalet’s server. They are not enforced by an on-chain program. Whether execution is switched on for this deployment is shown on the status page. Monalet does not build wash trading or volume tools, and nothing here promises a profitable trade.
Activity records
Every action is recorded: who did it, what happened, a one-sentence summary and the evidence behind it. For an agent’s action the summary states the decision and its stated basis; the model’s internal reasoning is neither stored nor shown.
- Stored by Monalet: the activity history, conversations, drafts, the credit statement, proposals and their checks. They live in Monalet’s database. The history and the statement accept additions only, which guards against accidental edits; it is not independent proof, and these records should not be called immutable.
- Verifiable on chain: a trade whose transaction was confirmed on Solana. Only those entries are marked “On chain”, and each links to the transaction.
What Monalet does not do
- It does not hold funds or keys, and it does not sign for you.
- It does not deploy tokens.
- It does not sell credits or take payments.
- It does not credit fee income: no fee mechanism is deployed.
- It does not enforce trading limits on chain.
- It does not post to X, and it never posts anywhere on its own.
- It does not give financial advice. Agents are AI and can be wrong.
The status page shows which of these capabilities are working on this deployment right now.