The complete guide to Pump.fun Volume Bots.

What a Pump.fun Volume Bot actually does

The marketing copy for most volume bots in 2024 was a single sentence: "we make your token's chart go up". That framing has aged badly because it described a tool that, at best, moved a single signal — on-chain trade volume — and ignored the four other signals that decide whether anyone notices.

A Pump.fun volume bot in 2026 is best understood as a signal coordinator. It runs a fleet of ephemeral wallets that touch a token from five different angles, simultaneously, in patterns the trending algorithm reads as "organic discovery". The five angles are:

• On-chain trade volume on the bonding curve — and later on the AMM if the curve graduates.
• Native chat density in the Pump.fun comment stream, in the languages the audience actually speaks.
• Watchlist velocity — the rate at which distinct accounts star the token in their Pump.fun watchlist.
• Holder spread — the on-chain count of distinct addresses holding any non-zero balance.
• Cross-DEX presence — whether the token shows up on aggregator surfaces (Jupiter, DexScreener) with non-trivial liquidity, not just the launchpad.

A bot that moves only the first signal is the cheap, easy-to-detect species. A bot that moves all five, in patterns shaped to the algorithm's sample windows, is what the rest of this guide is about.

Pump.fun bonding curve mechanics in plain words

Before talking about how to push a token on Pump.fun, you need a working mental model of the bonding curve itself. Pump.fun's curve is a constant-product invariant on a synthetic SOL/token pair. The curve has a fixed total supply (1 billion tokens) and a starting "virtual" SOL reserve that biases price discovery toward early buyers.

What this means in practice:

1. Every buy moves the price up non-linearly — late buyers pay sharply more SOL per token than early buyers.
2. Every sell moves the price down by the same constant-product math — the curve is symmetric.
3. The curve has a graduation cap at a fixed SOL reserve threshold (commonly cited as the ~$69k market-cap line). When the cap is hit, Pump.fun "burns" the curve and migrates liquidity to a Raydium AMM pool.
4. Until graduation, all trades happen on the curve — no AMM, no order book, no LP positions to manage.

The implication for a volume bot is non-trivial. Buy pressure on the curve is asymmetric: a 0.5 SOL buy at curve-position 5% does very different work than the same 0.5 SOL buy at curve-position 80%. A naive bot that fires equal-size buys at constant intervals will pay enormously over the odds at the top of the curve and barely move price near the bottom. A curve-aware bot scales trade size inversely to curve position so the price-per-trade impact stays even across the run.

The trending-board signal stack

Pump.fun's trending board is the prize most launches are actually competing for — placement there is the difference between 30 holders and 3,000 holders in the same hour. The board is not ranked by raw volume. It is ranked by a composite score that re-samples each token at intervals and weighs at least four observable signals:

• Recent volume velocity. Not cumulative volume — the derivative of volume over the last sample window. A token doing 5 SOL/min is ranked above a token that did 500 SOL three hours ago and nothing since.
• Distinct-buyer count. Twenty distinct wallets buying once each rank above one wallet buying twenty times. This is the signal that punishes the lazy bots that recycle a small wallet pool.
• Native chat activity. The Pump.fun chat for that token. Density, recency, and (we strongly suspect) language diversity all factor in. A chat that reads like a multilingual community is weighted above a chat that reads like a script firing the same template.
• Watchlist velocity. The rate of distinct accounts starring the token. This is the most under-exploited signal — most bots ignore it entirely. It is also the cheapest single move you can make to push board ranking.

The composite is sampled, decayed, and re-ranked on a cadence we estimate at 60–120 seconds during peak hours. The implication: a session that lands a 90-second pressure pulse across all four signals at the start of a sample window will outrank a session doing 5× the volume but only on the first signal.

Wallet rotation: depth, recycling, clustering

The single most predictive variable for "does this bot get its tokens flagged" is wallet pool depth. Every Solana on-chain forensics tool — Bubblemaps, Solscan, custom DEX-team dashboards — uses the same handful of clustering heuristics:

1. Funding-source clustering. Wallets all funded by the same upstream wallet within a tight time window cluster together.
2. Behavioral clustering. Wallets that fire the same trade size at the same cadence on the same tokens cluster, even without a shared funding source.
3. Reuse clustering. A wallet that appears across multiple suspicious launches is permanently flagged.

The defensive moves are mechanical:

• Pool depth. A pool of 10,000+ never-reused wallets means each launch pulls a fresh subset. No single launch's wallets appear in another launch's footprint.
• Funding fan-out. Sub-wallets are funded from a randomized fan-out tree, not a single hub. The on-chain graph reads as many small flows from many small sources, not one big distribution event.
• Trade-size jitter. Per-trade SOL amounts are sampled from a persona-specific distribution, not picked from a fixed set.
• Inter-trade timing jitter. Poisson-distributed waits between trades break the regular cadence forensics tools train on.
• Block-gap enforcement. No two sub-wallets in the pool fire trades on the same Solana block — kills the cheapest "two trades in one block ⇒ same actor" heuristic.

Anti-MEV: why mempool execution is a tax

If your bot routes trades through the public Solana mempool, every meaningfully sized buy is partly extracted by sandwich bots before it lands. The math is depressing once you write it down: the median sandwich on a 0.5 SOL buy on a thin Pump.fun curve in 2025 cost the buyer ~38 bps in slippage, every single trade.

For a 1,000 SOL session that fires ~12,000 trades, that is roughly 4–8 SOL extracted by MEV bots before any execution-quality discussion. It is a quiet tax that nobody on the marketing side of a bot business ever mentions.

The fix is not "be sneaky in the mempool". The fix is do not route through the mempool at all. Jito private relays accept signed bundles, route them straight to validator block-builders, and skip the public mempool entirely. Sandwich bots cannot see the trade because the trade was never broadcast.

The trade-off: bundles cost a priority tip, sampled from a configurable range. Tips that are too low miss block inclusion. Tips that are too high turn into the same MEV tax in a different costume. The right answer is a tip-tuning loop that auto-aligns to the current network priority-fee percentile, not a fixed range. We covered the implementation details in our deep-dive on Jito routing.

Pump.fun → Raydium migration handoff

When the bonding curve graduates, the same token continues life on a Raydium AMM pool — but with a discontinuity that breaks naive bots. The graduation is a single transaction in which the curve is closed, the SOL reserve is moved to a new Raydium pool, and the token's "venue address" effectively changes.

A bot that does not detect the graduation block will, in the seconds after, fire trades against an empty curve and lose them to errors, or worse, route to a stale liquidity assumption. Both outcomes look identical from the dashboard: the volume curve flatlines.

The defensive move is block-by-block detection. The bot subscribes to the token's curve account, watches for the graduation instruction, and the moment the instruction lands, re-points routing to the new Raydium pool — same session, no manual stop-and-restart. The user sees a continuous run; under the hood, the venue and the trade primitives both changed.

For tokens that do not graduate (the majority), the migration logic never fires and is a no-op. For tokens that do graduate (the ones that matter), it is the difference between a session that ends mid-pump and a session that rides the migration into the new market.

Curve presets explained (and when each one breaks)

Most bots offer a small library of named curves: Gradual, Burst, Stealth, Whale Pump. Each is a shape on the volume-over-time graph, and each has a use case where it works and a use case where it fails.

Gradual

Even pressure across the session window. Best for: tokens that need to look organically discovered — early-stage launches with weak existing community, where a sharp spike would read as obviously inorganic. Breaks when: the audience expects a launch event and the slow build looks unconvincing.

Burst

Front-loaded pressure with a sharp 8–15% spike to trigger trending sample windows. Best for: established communities pushing for visibility — the audience knows the launch time and a flat line would read as a failed launch. Breaks when: the burst is too sharp and trips heuristic spike-detectors that some screener boards now run.

Stealth

Distributed pressure that never spikes above natural-volume noise. Best for: accumulation runs where you specifically do not want screener attention. Breaks when: the run is large enough that aggregate volume gives the game away even without a spike.

Whale Pump

A few large persona buys at strategic curve points, supported by retail-persona infill. Best for: "whale rotation" narratives where the visible signal is "a whale is accumulating". Breaks when: the whale persona's wallet history is thin and on-chain analysts notice that "the whale" was created twenty minutes ago.

The cost model that actually makes sense

Three pricing models dominate the volume-bot market:

1. Subscription + per-trade gas + tips. The most common model. Looks cheap on the marketing page, ends up at 6–10% of session volume by the time gas, priority fees and "boost" prompts have been tallied.
2. Per-trade percentage. A flat % of each trade's SOL amount. Honest, but compounds painfully over the thousands of small trades a single session fires.
3. Flat % on session target volume. A single number — say 2% — on the volume the user asks the bot to generate. Every fee inside.

The flat-on-target model is the only one that aligns the bot's incentives with the operator's. Subscriptions reward the bot for collecting fees regardless of execution quality. Per-trade percentages reward the bot for firing more, smaller trades. Flat-on-target rewards the bot for hitting the volume target as efficiently as possible — which is what the operator actually wants.

Real volume vs. fake volume (what trending sees)

The phrase "fake volume" is doing a lot of work in this conversation. The trending algorithm does not see "fake" or "real". It sees signals it can or cannot distinguish from organic patterns. The relevant axis is not real vs. fake — it is distinguishable from the median organic launch vs. not.

The signals that distinguish a low-quality bot from organic activity:

• Trade-size mode (a bot that fires the same 0.1 SOL trade 5,000 times has a unimodal trade-size distribution; organic launches have a long tail).
• Inter-trade timing mode (a bot with a fixed 3-second cadence is trivial to flag).
• Wallet-pool depth (twenty distinct wallets generating a million SOL of volume is louder than a megaphone).
• Comment text similarity (a comment database with 50 strings is easy to fingerprint).
• Geographic / language uniformity (a chat that is 100% English on a token that is supposed to be a global launch is suspicious).

Conversely, the signals that don't distinguish a high-quality bot from organic activity at all: time-of-day, total volume, holder count growth, watchlist count growth. Aggregate stats look identical between a 10,000-wallet sophisticated session and an organic launch with 10,000 buyers. That is the entire point.

Eight common mistakes

1. Sizing the session by ego, not by curve position. A 5,000 SOL session on a token already past 70% curve position has nowhere to go and burns money. Match session size to runway.
2. Using a wallet pool below 500. Bubblemaps clusters this in minutes.
3. Routing through the public mempool. 4–8 SOL leaked to MEV per 1,000 SOL session.
4. Ignoring auto-favorites. Cheapest trending push you can make.
5. Single-language comments. Pattern-matchable, weakly weighted by the algorithm.
6. No migration handoff. Session dies the moment the curve graduates.
7. Buying back from your own deposit wallet. Visibly self-buying — instant credibility kill on Bubblemaps.
8. Holding the same wallet pool across multiple launches. Each new launch carries forward the cluster from the previous.

What to read next

If you got value out of this, the natural follow-ups are the two pillar articles that go deep on the harder parts: