Whale Token Sniper Bot Tutorial: Track Big Wallets on Base (2026)
9 min read
Whale sniping is one of the oldest meme-coin strategies that actually works. A small number of wallets on any launchpad consistently call entries better than average, and copying their flow with a small lag captures a slice of that edge. The hard part is plumbing — most launchpads do not expose pre-execution order intent, every copy pays gas, and most copy bots cost a subscription. THRYX gives you all three for free if you wire it up correctly.
What is a whale sniper bot?
An automated agent that does three things on a loop: watch a small set of high-volume wallets, decide which trades are worth copying, and submit a same-direction trade before the move plays out. The "sniper" part is speed. The "whale" part is selection — most platforms have a long tail of unprofitable wallets, and copying them indiscriminately is worse than coin flipping. A real whale-tracker filters on a wallet's historical PnL before it copies anything next.
Most public copy-trading bots run on Solana or Base by polling third-party APIs for trades. They cost real gas per copy, the data is rate-limited, and the lag between the whale's tx and the bot's tx is usually 5–15 seconds because the bot has to subscribe to a websocket, parse logs, classify the trade, and submit a new tx with its own gas tip. The same flow on THRYX is shorter because the launchpad backend already owns every step of that pipeline.
Why THRYX is the right venue for this
Three properties of THRYX make whale sniping cheaper and faster than running it on a generic L2.
- Pre-execution order flow. Every authenticated /api/tokens/:address/estimate call records intent before the trade fires — so a wallet hovering over the buy button shows up in the flow tracker before its on-chain swap lands. Exposed to your agent via get_recent_trades and get_trending.
- Gasless trades. Every copy is sponsored by the on-chain paymaster funded out of the protocol fee. Your bot never holds ETH, never pays gas, never tops up.
- No API keys. The MCP server is JWT-authed off your THRYX account. Sign in once, mint a token, hit the server.
Two paths to ship one
There are two ways to deploy a whale-sniper on THRYX. The choice depends on whether you want to write code.
| Path | Where it runs | Custom code? | Best for |
|---|---|---|---|
| A — On-platform autotrader | thryx.fun/autotrader | No, just a system prompt | Most users — fastest to ship |
| B — External MCP client | Your machine or VPS | Yes — Python or Node | Devs who want a custom data layer |
Both paths use the same gasless settlement and the same pre-execution flow data. Both inherit the same server-side caps because the MCP server is a thin RPC over the same /api/trade/buy and /api/trade/sell routes — not a privileged backdoor.
Path A — configure the on-platform autotrader
Open thryx.fun/autotrader. The page is a single form: cadence, max position size, daily loss cap, system prompt, and enable toggle. The agent runs Llama 3.3 70B via Groq, ticks on your cadence, and decides each cycle whether to buy, sell, broadcast, react, or skip — constrained by seven read-only tools and five action verbs.
Recommended starting caps for a brand-new whale-tracker bot:
- Cadence: 2 minutes. Often enough to catch a whale entry, slow enough not to thrash.
- Max position size: 0.0002 ETH (~$0.50). Server-side ceiling regardless of what the model wants.
- Daily loss cap: 0.001 ETH (~$2.50). Soft guideline shown in prompt; the agent self-throttles.
- Group chat enabled: off until you trust the agent's broadcasts.
- Mode: live (paper was retired 2026-04-28).
These caps are deliberately tiny. The point is to watch the agent's reasoning for the first dozen cycles and tune the prompt before raising real money at it.
A working whale-sniper system prompt
The prompt below is a complete starting point. Paste it into the system_prompt field on /autotrader. It is structured around the tools the agent actually has and the conviction tiers the platform enforces.
You are a whale-tracking sniper agent on THRYX, a gasless meme-coin launchpad on Base.
Goal: detect when a high-volume wallet enters a token, copy the entry with a small position, exit on either +20% or after 30 minutes flat.
Process every cycle:
1. Call get_recent_trades. Identify any trader who appears 3+ times in the last hour with positive sized buys, OR who shows a single buy >= 0.005 ETH.
2. For each candidate token, call get_token_details. Skip if: graduated == true, age < 60 seconds (anti-sniper hook charges 80%), or feeBps unknown.
3. Compute round-trip floor: 4.5% on tokens with feeBps == 100, 3.5% on tokens with feeBps == 50. Reject any plan whose target return is below 2x the floor.
4. If a candidate passes, propose_trade with side="buy", conviction tiered as:
- low (20% of cap) for one-off whale buy
- med (50% of cap) for repeated whale buys in 10 min
- high (100% of cap) only if 3+ whale wallets buy the same token in 30 min
5. For positions you already hold, call get_my_holdings. Sell when:
- unrealized PnL >= +20%
- position age >= 30 minutes regardless of PnL
- the same whale that triggered entry has sold (visible in get_recent_trades)
6. Use remember to log every closed trade with the trigger wallet and outcome. This is your edge.
7. Skip cycles where no candidate passes. Do NOT trade for the sake of trading.
Hard rules:
- Never propose more than 3 buys per cycle.
- Never buy a token already in get_my_holdings if avg cost is within 5% of current.
- Never sell at minOut=0 — refuse and retry next cycle.
- Treat any token with token.feeBps undefined as a hard skip.
Done condition: emit done() when the cycle has either placed a trade or affirmatively decided not to.A few details worth flagging. The 60-second skip on freshly-launched V4-native tokens is the anti-sniper hook deciding the trade for you — at second one, the hook charges 80%, decaying to 1% over 60 seconds. Buying inside that window is a structural loss. The conviction tiers map directly to the 20/50/100% sizing the platform enforces. The remember() calls accumulate into the per-user user_agent_memory table and reload into context on every subsequent cycle — that is how the bot builds a per-whale reputation score without you writing any database code.
The agent's tool surface
Every cycle, the on-platform agent has access to:
| Tool | Returns | Use in whale-sniping |
|---|---|---|
| get_recent_trades | Recent buys and sells across the platform | Primary signal — whale entries land here within seconds |
| get_trending | Top tokens by recent volume | Confirm the whale is moving a token with breadth, not just one wallet |
| get_graduating | Tokens approaching V4 graduation | Bonus signal — whale + near-graduation = high-conviction setup |
| get_token_details | Spot price, depth, feeBps, holder count | Compute round-trip floor and reject thin tokens |
| get_my_holdings | Current positions with cost basis and unrealized PnL | Manage exits |
| get_my_eth_balance | WETH balance available to spend | Budget check |
| get_my_memory | The agent's own past lessons and per-whale notes | Accumulated edge across cycles |
| propose_trade | Submit buy or sell with conviction tier | Action verb — only path to actually trade |
| remember | Persist a lesson, preference, or note | Build per-whale reputation over time |
| skip / done | End the cycle with no trade | Required to terminate cleanly |
Path B — build an external MCP bot
For a bot that runs outside the platform — on your laptop, a VPS, or a CI runner — point an MCP-compatible client at the THRYX MCP server. Auth is a short-lived JWT minted from your THRYX session. Same tools, same gasless settlement.
The minimum viable external whale-tracker is a 30-line loop: poll the recent-trades endpoint, filter on a whale-list you maintain locally, call estimateSwap to compute minOut at 95%, post to /api/trade/buy with the JWT in the Authorization header. The relay forwards to the Diamond at 0x2F77b40c124645d25782CfBdfB1f54C1d76f2cCe and the paymaster sponsors the gas. The trade-off versus Path A: you give up the LLM reasoning layer and have to write the decision logic yourself.
Caps, safeguards, and the round-trip floor
The platform enforces hard server-side caps on every agent-driven trade regardless of what the LLM produces:
- MAX_TRADES_PER_CYCLE = 5 on-chain trades per cycle.
- PER_CYCLE_BUDGET_MS = 90,000 — cycle hard-stops at 90 seconds.
- Anti-loop watchdog — 4 consecutive identical plans abort the cycle.
- Pre-flight gates: sell-no-holdings, eth-balance check, partial-token detection, same-cycle dedupe.
- Per-trade max position size from your /autotrader config.
On top of the caps, the round-trip floor is the math you cannot bypass. Every buy-then-sell pays one swap fee on each leg. Net floor is 4.5% on new V4-native launches (1% per leg) and 3.5% on the ~600 legacy tokens grandfathered at 50 bps via the v2.14 setTokenFeeBatch migration. A whale-tracker that does not target at least 2x the floor on every entry is going to bleed slowly.
Reward stacking — why the bot pays you to run it
Every trade your bot submits earns THRYX rewards on top of any PnL. The post-2026-05-02 schedule:
| Trade size | THRYX reward | Daily cap |
|---|---|---|
| >= 0.0001 ETH (base) | 10,000 | 15 trades/day |
| >= 0.0005 ETH | 40,000 | (in 15 cap) |
| >= 0.001 ETH | 100,000 | (in 15 cap) |
| >= 0.005 ETH | 500,000 | (in 15 cap) |
| >= 0.02 ETH | 1,500,000 | (in 15 cap) |
A whale-tracker at the recommended 0.0002 ETH cap collects the 10K base reward per trade. Streak bonuses stack: 25K THRYX/day from day 2, 100K/day from day 7, 500K/day from day 30. The bot does not need to know about rewards — they accrue automatically and are claimable through the ClaimFacet whenever you sweep.
What this won't do
A few honest limitations. THRYX is a small platform — 129 users at the time of writing — so "whale" here means the top handful of wallets by lifetime volume, not a Binance market maker. The signal is real but thin. The agent will frequently sit through cycles where nothing meaningful happened, and that is the right behavior. A bot that trades on every cycle to look busy is a bot that is bleeding the round-trip floor on noise. Skipping is the default. A prompt tuned to the current top-five wallets will also drift as those wallets change behavior or get replaced — plan to re-read the agent's remember() history every couple of weeks and update the prompt with what is and is not working.
Frequently asked
- Do I need ETH in my wallet to run this bot?
- No. THRYX is gasless end to end. The on-chain paymaster sponsors the gas on every trade your bot submits, funded out of the 30% protocol cut on each swap fee. Your wallet ETH balance does not move on a buy or a sell. You do need WETH to spend on entries, but no ETH for fees.
- How fast does the bot react to a whale entry?
- On Path A (the on-platform autotrader) the cycle cadence is your config minimum. The default minimum is 1 minute. On Path B (external MCP client) you control the polling cadence and can react in seconds. Either way, the round-trip from observed whale tx to your copy tx is bounded by your loop, not by the chain — Base has 2-second blocks and the relay submits within one block.
- Can the bot lose money?
- Yes. Every individual trade has variance, and a whale-tracker with a bad prompt or no filter on whale quality will lose to the round-trip floor. Start with the smallest possible caps (0.0002 ETH max position, 0.001 ETH daily loss), watch the agent's reasoning in the recent-decisions list on /autotrader, and only raise caps once the running PnL on tiny size is positive.
- Where do I find which wallets are whales on THRYX?
- Open thryx.fun/leaderboard for the on-platform leaderboard, or call get_recent_trades from the agent and let the LLM derive it from observed flow. The platform does not publish a fixed whale list — that is part of the edge: you build your own from the cycle history.
- Does this work on graduated tokens too?
- Yes. Graduated tokens trade through the same Diamond via the V4 SwapFacet branch — same gasless flow, same fee math. The only difference is estimateSwap returns 0 for graduated tokens by design, so the bot derives minOut from the cached priceUsd at 95% slippage instead.
Open the THRYX AutoTrader and start your bot
Related Posts
- Inside the THRYX AutoTrader — The graduation-sniper companion to this guide.
- Claude Code Crypto Trading Tutorial — Drive the same MCP server from your terminal.
- Building with THRYX: MCP + CLI for AI Agents — Full integration reference.
- How Does Token Graduation Work? — The mechanic the second-best whale signal trades around.