Skip to main content
A token can move ten percent in a minute, and the candle alone never tells you why. Most of the time the real cause is a handful of large trades clearing a thin order book. A whale transaction tracker watches the trades themselves, so a six figure buy or a sudden wave of selling shows up the moment it happens, not after the move.

TL;DR

  • Read a token’s normal trading stats so the threshold is not a guess
  • Pull only the trades that clear that threshold
  • Classify each print as a buy or a sell
  • Poll for new prints on a short loop so alerts fire near the moment a whale trade lands
Whale transaction tracker pipeline showing four stages from baseline volume to a live large trade alert All requests share the base URL https://public-api.birdeye.so, authenticate with the X-API-KEY header, and select the network with x-chain: solana.
The trades endpoint that drives detection, classification, and polling is Solana only, so this tracker targets Solana. The baseline endpoint in Step 1 supports other chains if you need aggregate stats elsewhere.

​
The four stage pipeline

1

Size a threshold from the token's own baseline

A whale is relative. A 20,000tradeisenormousonatokenwith20,000 trade is enormous on a token with 50,000 of daily volume and barely noticeable on one with $50 million, so the first call is not detection, it is context.Endpoint: GET /defi/v3/token/trade-data/single
This endpoint uses address, not token_address, the opposite of the endpoint in the next step. Size every threshold from the _usd fields, never the raw token quantity fields: volume_buy_24h is a token amount, while volume_buy_24h_usd is the dollar figure, and buy_24h/sell_24h are trade counts, not volume at all.
A simple starting heuristic: set min_volume near 0.5 to 1 percent of volume_1h_usd, then tighten or loosen it based on how many alerts you actually act on. Revisit the number whenever volume shifts, since a fixed dollar figure quietly becomes too loose or too tight as the token’s activity changes.
2

Pull only the trades above that threshold

Most tokens generate thousands of small trades for every one that matters. Filtering for size up front means the tracker only ever looks at prints worth caring about.Endpoint: GET /defi/v3/token/txs-by-volume
volume_type has no default. A missing value returns a 400 error rather than falling back to a sensible choice. Read volume_usd for sizing, not the raw volume field, since its denomination is not fixed and sometimes reports the paired token’s amount instead of the monitored token’s.
3

Classify each print as a buy or a sell

A whale buying and a whale selling look identical on a line chart but mean opposite things for where price goes next.This is the same endpoint as Step 2, used two ways. Either read each print’s side field from a combined pull, or issue two calls with tx_type=buy and tx_type=sell to get each direction pre split.
The convention is relative to the token you query: a print labeled buy sits in the to leg with a positive ui_change_amount, and a sell sits in from with a negative amount. Query the specific token you are monitoring, not a quote token like SOL, or the labels stop reading correctly, since the trades returned would span every pair that token touches.
Sum volume_usd per direction over a rolling window to get net pressure rather than a raw trade count. A token can show ten buys and two sells and still be net negative if the two sells are large enough.
4

Poll for new prints on a short loop

A whale tracker that only answers when asked is a research tool, not an alert system. This stage turns the same endpoint into a live feed.
Set after_time to the block_unix_time of the last print you handled, and dedupe on the composite key tx_hash plus ins_index plus inner_ins_index. One transaction can produce several trade legs sharing a single hash but differing by index, so deduping on tx_hash alone drops real legs rather than just true duplicates.
A polling interval of two to five seconds catches most whale activity without straining your credit budget. Going much below roughly one second rarely buys meaningful extra warning, since a trade still has to confirm on chain before it shows up in the response.

​
Watch your credit budget

A whale tracker polling on a tight loop across several tokens can consume credits faster than a one off script. Endpoint: GET /utils/v1/credits
Use the result to tune your polling interval and the number of tokens you watch, so coverage and cost stay in balance.

​
Before you ship

  • Every threshold is sized from the _usd fields, never the raw token quantity fields.
  • volume_type=usd is always set explicitly, since a missing value returns a 400 error.
  • The specific token being monitored is queried, not a quote token like SOL, so buy and sell labels read correctly.
  • volume_usd is summed per direction to get net pressure instead of counting trades.
  • Dedupe runs on the composite key, not tx_hash alone, so multi leg transactions are not collapsed.

​
FAQ

There is no universal dollar figure, since a large trade on one token is routine on another. Set the threshold relative to the token’s own baseline, using volume_1h_usd or volume_24h_usd, rather than a single number applied across every token you watch.
Read the print relative to the token you queried. The monitored token sitting in the to leg with a positive ui_change_amount is a buy, and the same token sitting in from with a negative amount is a sell. Query the specific token you care about rather than a quote token, or the labels stop reading correctly.
Yes, the same endpoint that powers live polling also answers a historical question. Set before_time and after_time to bound a past window instead of polling forward. This is useful for reviewing what drove a price move after the fact, or for backtesting a threshold before committing to it live.
Yes, by running the same four stages per token rather than building anything new. Keep each token’s threshold tied to its own baseline instead of reusing one number across very different tokens, since a single fixed floor floods you with noise on a quiet token and misses everything on an active one.
That closes the loop. A market wide stream of activity becomes a precise event: a trade large enough to matter, on the token you are watching, with a direction attached.