Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.
Token due diligence
token_due_diligence scores a Solana token on who holds it, not only on its contract flags. Every field is computed from chain data we read ourselves (token accounts, signature history, funding transfers) plus our own tracked wallet history. No third-party labels are scraped.
- API:
GET /api/crypto/due-diligence?address=<mint>andGET /api/crypto/funding-graph?wallet=<address> - MCP tool:
token_due_diligence(input{ mint }) - Token page: the "Holder due diligence" panel on
/threerenders the fields and the two-hop funding force graph - Cache: one report per mint for 300 seconds. Only successful reports are cached.
Method
- Find the launch: the oldest transaction of the mint's Metaplex metadata account (falling back to the mint's own history). Its slot is the launch slot and its fee payer is the creator.
- Read the largest token accounts (up to
analyze_top) and resolve each to its owner. Program-owned holders (pools, vaults) are not wallets and are excluded from wallet ratios. - For each wallet holder read its first receipt of the token, its first-ever transaction, and who funded it, up to two hops back.
- Classify holders and aggregate by share of supply.
Thresholds
Every threshold lives in DD_THRESHOLDS in api/_lib/token-due-diligence.js and is returned in each report.
| Key | Value | Meaning |
|---|---|---|
analyze_top |
20 | Largest token accounts analysed per mint (the chain returns at most 20). |
sniper_slots |
5 | A holder is a sniper when its first receipt landed within this many slots of launch. |
bundler_min_group |
2 | Holders sharing a funder form a bundle from this group size up. |
same_slot_min_group |
3 | Same-launch-slot grouping needs this many holders, because organic buyers share busy slots. |
funding_window_hours |
6 | Wallets funded by one source within this window of each other form a cluster. |
fresh_wallet_max_age_hours |
24 | A wallet is fresh when its first-ever transaction is younger than this at first receipt. |
smart_money_min_score |
65 | wallet_reputation.smart_money_score at or above this bar. |
service_funder_min_sigs |
2000 | A funder with this many signatures (or an incomplete history) is a service such as an exchange or router, not a coordinator, and never links holders. |
min_evidence_ratio |
0.5 | A field is reported only when this share of analysed wallet holders yielded the evidence it needs. Below it the field is null with a "withheld" note, never a low number. |
Definitions
insider_pct
Insider hold rate. Percent of supply held by the creator wallet and by holders linked to it: funded by the creator, funding the creator, or funded from the same non-service source within two hops.
sniper_pct
Sniper share. Percent of supply held by wallets whose first receipt of the token landed within sniper_slots slots of the mint's first transaction. The creator is excluded (counted as an insider).
bundler_pct
Bundler exposure. Percent of supply held by wallets in a funding cluster (two or more holders funded by one non-service source within funding_window_hours, directly or through a shared second hop) or in a launch slot where same_slot_min_group or more holders first received the token.
fresh_wallet_pct
Fresh-wallet share. Percent of supply held by wallets whose first-ever transaction was under fresh_wallet_max_age_hours before they first received the token.
fresh_wallet_ratio
Fresh wallets divided by analysed wallet holders (0 to 1). Pools and other program-owned holders are excluded from both sides.
smart_money_holders
Count of analysed holders whose own realized history in our tracked data (wallet_reputation.smart_money_score) is at least smart_money_min_score.
smart_money_pct
Percent of supply held by the smart-money holders above.
Funding graph
traceFundingGraph(wallet) follows who funded a wallet's first transaction, then who funded that funder, two hops in total, and returns nodes and edges. The token page renders it as a force graph with the creator, flagged holders and service funders distinguished in the legend.
Guards
Four per-agent trade limits refuse a buy when the report exceeds them. All default to off (null).
| Limit | Refuses a buy when |
|---|---|
max_insider_pct |
insider_pct is above the limit |
max_sniper_pct |
sniper_pct is above the limit |
max_bundler_pct |
bundler_pct is above the limit |
max_fresh_wallet_pct |
fresh_wallet_pct is above the limit |
Set them through the agent trade limits endpoint. The report is loaded only when a limit is set, and applies to mainnet buys. If a limit is set but the report or the field cannot be read, the guard fails closed with due_diligence_unavailable rather than letting the trade through.
Coverage caveat
Reads use the public RPC failover chain. Some lanes prune history, return empty lists or rate-limit, so a report states its coverage (coverage) and withholds any field below min_evidence_ratio. A withheld field means "not enough evidence", not "safe". Retry later or configure an archival RPC lane for fuller coverage.