[KongchangAI]
Product

Sniff

A Meme coin provenance agent built on a graph database that reveals the real parties associated with a token by analyzing graphs of wallet funding relationships and token creation relationships.

Timeline (last 90 days)

Oct 18

The QuantBase team built an agent called Sniff at the WeMakeDevs × FalkorDB Graph Hacks hackathon for memecoin tracing.

Unverified50%
Oct 18

Sniff's data snapshot retains 632,339 tokens, 424,007 wallets and about 1.5 million relationships.

Unverified50%
Oct 18

The FalkorDB bulk loader completed the Sniff data import in 30 seconds, and the graph uses only 521MB of memory.

Unverified50%
Oct 18

In Sniff's graph, any two random tokens are connected within four hops, because exchange hot wallets connect large numbers of wallets as funders.

Unverified50%
Oct 18

The team flags addresses that funded more than 200 other wallets as hubs, identifying 89 hubs in total.

Unverified50%
Oct 18

Sniff sets a hard rule: no query may pass through a hub address.

Unverified50%
Oct 18

After excluding hubs, FalkorDB's algo.WCC algorithm found 6,556 weakly connected components (rings) in the Sniff graph.

Unverified50%
Oct 18

Of 3,900 multi-creator clusters, 3,824 fall entirely within a single WCC, showing that graph clustering and the old method cross-validate with high agreement.

Unverified50%
Oct 18

Sniff's AI (Claude) only decides which of nine read-only tools to call (at most eight times) and never writes Cypher queries itself.

Unverified50%
Oct 18

Sniff's citation checker deletes entire sentences that are not backed by a valid evidence ID.

Unverified50%

4 more timeline events

All Facts (14)

Unverified

Sniff never says "safe" or "scam", never gives buy or sell advice, and every answer ends with "history, not prediction".

50%
Unverified

Sniff's data snapshot retains 632,339 tokens, 424,007 wallets and about 1.5 million relationships.

50%
Unverified

The FalkorDB bulk loader completed the Sniff data import in 30 seconds, and the graph uses only 521MB of memory.

50%
Unverified

In Sniff's graph, any two random tokens are connected within four hops, because exchange hot wallets connect large numbers of wallets as funders.

50%
Unverified

The team flags addresses that funded more than 200 other wallets as hubs, identifying 89 hubs in total.

50%
Unverified

Sniff sets a hard rule: no query may pass through a hub address.

50%
Unverified

After excluding hubs, FalkorDB's algo.WCC algorithm found 6,556 weakly connected components (rings) in the Sniff graph.

50%
Unverified

Of 3,900 multi-creator clusters, 3,824 fall entirely within a single WCC, showing that graph clustering and the old method cross-validate with high agreement.

50%
Unverified

Sniff's AI (Claude) only decides which of nine read-only tools to call (at most eight times) and never writes Cypher queries itself.

50%
Unverified

Sniff's citation checker deletes entire sentences that are not backed by a valid evidence ID.

50%
Unverified

Tested on 23,408 graduated tokens, adding information that anyone could see in the first few minutes changed the predictive accuracy of Sniff's graph signal by within ±0.01, with no measurable gain.

50%
Unverified

On its own, a token whose social account has already been used by other creators is 18 percentage points less likely to still be alive one hour after graduation.

50%
Unverified

Sniff's nine tools are provided through an MCP endpoint, with the command `claude mcp add --transport http sniff https://sniff.quantbase.live/mcp`.

50%
Unverified

The QuantBase team built an agent called Sniff at the WeMakeDevs × FalkorDB Graph Hacks hackathon for memecoin tracing.

50%

Source Articles