The agent
An AI agent that scores every Base launch — in real time.
Hundreds of new tokens deploy on Base every day. SHADE reads them so traders don't have to: a single 0–100 composite plus a discrete set of risk flags, in under 200ms.
Three layers. One pipeline from launchpad event to your inbox.
Ingestion & scoring
An Erigon archive node with a custom eth_subscribe filter on the Clanker, Flaunch, Bankr, and Zora deployment factories feeds a Kafka bus partitioned by launchpad with a 7-day retention window. Postgres enrichment joins each deployment against indexed historical state in under 200ms p95.
- Clanker · Flaunch · Bankr · Zora
- Kafka — 7-day replay window
- Postgres enrichment <200ms p95
- Farcaster Hub cast firehose
Privacy primitives
Four production primitives compose at four different leak points: Semaphore for subscription, XMTP for delivery, Phala TEE for inference, Railgun for trade routing. Each addresses a separate adversary in the threat model.
- Semaphore · BN254 / Groth16
- XMTP · X3DH + Double Ratchet
- Phala · SGX / SEV-SNP
- Railgun · shielded UTXO
Delivery & action
Multi-channel delivery — XMTP inbox, zk-gated Telegram bot, encrypted Farcaster DM. Builders with a premium credential receive an authenticated WebSocket stream with per-credential token-bucket rate limits. Optional pre-trade risk check for Bankr.
- XMTP inbox
- zk-gated Telegram
- Encrypted Farcaster DM
- Builder WebSocket API
On-chain inputs
Read every deployment.
- Liquidity
- Pool depth, lock contract verification, Uniswap v3/v4 state.
- Holders
- Top-10 share, Gini coefficient, Herfindahl–Hirschman index.
- Deployer
- Prior deploys via the same EOA, prior rugs by graph traversal of funding paths, ENS / Farcaster ID linkage.
- Bytecode
- Pattern matching against known templates and bad-actor opcodes via a Bloom-filter index.
Off-chain inputs
Read every signal.
- Cast volume
- Farcaster replies, recasts, and unique inflows in the first 60 seconds after deploy.
- Caster reputation
- Neynar reputation score and historical hit-rate of the casting account.
- X mentions
- X mentions weighted by follower-quality score.
Composite score
S(τ) = wd·D + wh·H + wc·C + wv·V + wb·B
A weighted blend of five sub-scores. Weights recalibrate weekly via gradient-boosted regression against realized 7-day performance. Methodology, weights, and historical hit-rate are public.
Deployer reputation
Holder distribution & liquidity health
Caster reputation & hit-rate
Engagement velocity
Contract pattern analysis
Sub-scores
- D(τ)
- Deployer reputation
- Hit-rate of the deploying wallet across prior launches, weighted by recency and volume. Recency decay constant λ = ln(2)/30 days (30-day half-life); a launch counts as a hit if peak ≥ 5× initial.
- H(τ)
- Holder & liquidity health
- (1 − h₁₀) · min(ρ/0.05, 1) · 𝟙[locked]. h₁₀ is top-10 holder share; ρ is liquidity-to-FDV ratio; locked is the lock-contract verification flag.
- C(τ)
- Caster reputation
- Caster mentions weighted by Neynar reputation score and the caster's historical hit-rate.
- V(τ)
- Engagement velocity
- Slope of replies, recasts, and unique inflows in the first 60 seconds after deploy.
- B(τ)
- Bytecode pattern
- Binary discount for known-bad bytecode patterns: honeypot, mint backdoor, fee-on-transfer trap.