The Shibalytics field guide
A hosted, multichain AI trading agent you talk to in plain English. Here's exactly what it reads, how it decides, how it executes, and the one rule it never breaks: the chain keeps the score, and every number traces to something real.
What Shibalytics is
Shibalytics is a trading agent for on-chain markets. It does three things a chart can't: it surfaces the rare launches worth looking at, it fuses deep on-chain and social intelligence about them, and it executes — you talk to it like a person and it acts, from a wallet only you control. Then it grades every call against what actually happened.
It's built for the trader tired of two bad options: noisy alert bots that ping a hundred times a day, and black-box "signal" groups that show you their winners and hide their losses. Shibalytics is quiet, transparent, and hosted — nothing to set up, no keys to wire. You just talk, and it hunts.
What it is not. Not a signal-selling group, not a custodial exchange, not a price oracle. It doesn't promise a token will pump. It tells you what's verifiably true about who's buying, who built it, and who's being paid to talk about it — and then, on your say-so, it trades.
The trading harness
Most "AI agents" are a generic chat model bolted onto a few crypto APIs. Shibalytics is built on a rebuilt Hermes agent — an agent core re-engineered and tuned for one job: reading markets and acting on them. We host it and run it 24/7. You never touch an API key, a model config, or infrastructure. You talk; it hunts.
| Why a purpose-built harness | What it means for you |
|---|---|
| Tuned on outcomes | The harness is sharpened on measured wins and losses — the evidence loop is its training signal, not a frozen generic model. |
| Tool-native | It reasons directly over live on-chain and social evidence and calls execution tools itself — no brittle glue between "thinking" and "doing". |
| Hosted & always on | We run it around the clock. It hunts while you sleep; you just check in and steer. |
| No setup | No keys, no model choice, no config. Open it and go. |
Why hosted, not bring-your-own-key. A trading brain should be tuned on trading outcomes and screened for safety before it ever touches your wallet — not a raw general model you wire up yourself. Hosting lets us make it fast, safe, and sharper every week.
How the agent thinks
Every run is the same six-step loop. Four steps are our deterministic engine, one is the harness, and the last one closes the feedback loop:
- Surface engine — the backtested filter finds a candidate.
- Fuse engine — on-chain, social, and incentive signals are assembled into one evidence bundle.
- Reason harness — the harness scores the evidence against the rubric and returns a verdict with receipts.
- Execute you + engine — you approve (or pre-authorize within guardrails) and it trades from your non-custodial wallet.
- Measure engine — the call is auto-labeled against measured price outcomes.
- Learn engine — losing calls are dissected and the lessons tighten the harness.
The backtested algorithmic filter
Every day, tens of thousands of tokens go live across the chains Shibalytics watches — and the overwhelming majority are dead on arrival or engineered to take your money. The filter doesn't read charts or chase mentions. It records the opening minutes of virtually every launch, around the clock: who deployed it, who bought first, where those wallets got their funds, how supply spread out, and how each participant has behaved across every token they've ever touched.
Each wallet gets a profile built from 100+ on-chain measurements — deployer track record, holder concentration, entry timing, funding trails, holding discipline — so the system can tell a disciplined accumulator from a bot that flips in ninety seconds.
The filters are built backwards from outcomes. When a token produces an outsized run, we dissect its launch footprint and replay that exact signature against months of archived launches. If it would have flagged the winners and the scams, it isn't a filter yet; we keep stacking conditions until the historical rugs fall out and the runners remain.
On-chain metas rot, so filter maintenance is the product: every week the period's top performers are studied, stale conditions retired, and new ones promoted. The candidate feed is the output of that living loop, not a static ruleset frozen at launch.
Deliberately quiet. A surfaced token has already cleared dozens of independent checks before it reaches you — with wallet-level receipts so you can check our work. A handful of defensible candidates a day, not a hundred guesses.
On-chain intelligence
Once a candidate is surfaced, the agent needs to know who is actually in it. A raw buy feed is noise; a classified one is intent.
Wallet archetypes
| Archetype | Why it matters |
|---|---|
| Fresh | Newly funded wallets buying with size — often insiders or informed money on clean addresses. |
| Dormant waking | Long-inactive wallets suddenly buying — classic "smart old money", gated by that wallet's historical P&L. |
| Full-ported | A wallet that put its entire balance into one buy — maximum conviction. |
| Bundler / sniper | Block-0/1 coordinated buys — insider bundle exposure at entry, and a mean-reversion signal after they flush. |
| Leaderboard-profitable | Wallets with a measured win rate — weight the entry by who's already in. |
Beyond the buyers
- Smart money & holder quality — top-holder concentration, PnL, and the suspicious/fresh/sniper/bundler mix.
- Deployer forensics — serial-rugger bytecode fingerprints, deployer dormancy, hidden and MEV-bundled launches, pre-approvals before trading opens.
- Supply & exit tracking — how controlled the supply is, and — the part most tools skip — tracked cohorts selling, so distribution is caught as it starts.
- Contract safety — honeypot and sell-path simulation, so "can I actually sell this" is answered before your money finds out.
Social intelligence
On-chain tells you who's buying; social tells you why the crowd might follow. Shibalytics reads the narrative from multiple angles and weighs real virality over the manufactured kind.
- Narrative & mentions — who's actively posting the contract, breadth of unique voices, engagement-ranked context — not a raw count.
- Organic vs. crypto-native split — traction that escapes crypto-Twitter into mainstream communities weighs far more than in-crowd shilling.
- Deployer / handle credibility — account age, following, and history of the linked social handle as a cheap scam prior.
- Off-chain footprint — video, forum, and code-repository signals for narratives that originate outside the trenches.
Raw mention volume is gameable — entire communities are now paid to post. That's why social never stands alone here; it's cross-checked against the chain and the incentive radar below.
CT intelligence & the incentive radar
Crypto Twitter moves markets — but a large share of it is bought. This is the layer that prices that reality.
When investigators leaked one promotion campaign's books in 2025, 200+ influencers had price sheets running from about $50 to $60,000 per post, and fewer than five disclosed a single paid ad — payments made to wallets with on-chain receipts (ZachXBT, via The Block). Undisclosed touting is often literally illegal — the SEC fined one celebrity $1.26M for a single undisclosed promo (SEC). Shibalytics is built to detect that behaviour, not just measure the noise.
The tell is time ordering
Organic discovery: price and buys follow the post. Incentivized promotion: insider buys precede the post, and the promoter's wallet exits into the engagement peak. That asymmetry is detectable.
| Signal | What it surfaces |
|---|---|
| Incentive Score | Share of mention volume from known-paid or low-credibility handles; synchronized-burst / bot-amplification detection; deployer-linked supply behind the hype. |
| Caller Integrity | Per-KOL track record: do their calls systematically precede their own wallet's exits? |
| Coordination detection | Many low-quality accounts posting the same contract inside a tight window; same-day account creation; near-identical text. |
| Paid-alpha awareness | Weighing calls from paid groups as mercenary, front-runnable flow — not conviction. |
Confirm, fade, or flag
- Confirm — rising volume from credible, unpaid voices following organic on-chain accumulation. Social confirms the chain → conviction up.
- Fade — volume dominated by known-paid handles or a synchronized burst, with KOL wallets distributing into the spike.
- Flag — ambiguous provenance. Cap size, demand more on-chain confirmation, and surface the receipts so you decide.
Honesty note. Incentive scores are probabilistic inferences from public data — signals, not accusations. Detection covers public posts and monitored channels, not private group internals.
How a call is made
The engine assembles every layer above into one structured evidence bundle and hands it to the harness with the rubric. The harness returns a strict, machine-readable verdict:
# evidence in → hermes harness → verdict out { "token": { "symbol": "WOOF", "chain": "ethereum" }, "score": 74, "call": "buy", // buy | watch | pass "onchain": "3 dormant wallets woke; 2 CEX-funded fresh buyers; deployer clean", "social": "rising, 61% unpaid handles, organic reddit traction", "incentive": { "score": "low", "coordinated_burst": false }, "reasoning": "Accumulation precedes the social; hype isn't bought.", "receipts": [ "0x…wallet", "0x…deployer", "tweet_ids" ] }
The rubric encodes hard-won doctrine — narrative quality and timing lead, on-chain security is context not a headline, a strong narrative on the wrong chain is a red flag, and paid-alpha strength never overrides a weak thesis. Every verdict ships with its receipts, so you (and the harness, next time) can check the work.
The evidence loop & calibration
This is what separates Shibalytics from a signal group: every call is graded against measured price outcomes, mechanically, at fixed horizons — wins and losses alike.
Ran hard and held. The thesis played out.
Ran, then round-tripped — the pump-and-dump pattern, labeled for what it was.
Went the wrong way. Gets a post-mortem: which specific signal misled us?
Nothing conclusive yet, or a young token with little history — the agent says so instead of guessing.
Losing calls feed a learning pass that asks one question — which signal was misleading? — and turns recurring failure modes into concrete rules the harness inherits. The edge is the loop, not a frozen model.
Watch it think. Score-vs-outcome calibration, the miss board (candidates it passed on that ran), and per-signal accuracy are all surfaced. We show you where the agent is wrong, not just where it's right.
Steering your agent
You're not locked to the engine's read. A live context feed lets you tell the agent what you know — "the AI-agent meta is hot this week," "fade anything political" — as weighted, expiring directives that ride along with every decision. Your judgment and the machine's evidence meet in the same place.
Talking to it & execution
You interact in plain English. Tell the agent what you want and it handles the on-chain mechanics — no dashboards, no ten tabs of tooling.
"Buy half an ETH of whatever the filter loves this morning, but skip anything the shill radar flags. Set a stop at −18%."
Order types
| Capability | What it does |
|---|---|
| Market swaps | Buy/sell now at the best route the agent finds. |
| Limit orders | Fill only at your price. |
| Stop-loss | Auto-exit if a position breaks your floor. |
| Take-profit ladders | Scale out in steps as a position runs. |
| DCA / TWAP | Accumulate or unwind over time instead of all at once. |
| Bridging | Move value across supported chains as part of a trade. |
| Cross-asset triggers | Act on one asset when another moves — "sell my alts if ETH breaks $3k". |
| Signal triggers | Fire on on-chain evidence, not just price — "sell if the deployer wallet moves", "exit if the filter score falls below 70", "buy if smart money starts accumulating". The signals that front-run a dump become your stop. |
Scheduled strategies
Hand the agent a standing brief once and it runs the whole loop on a schedule while you're offline — scan, filter, size, execute, log. For example: "Every morning, scan new Ethereum launches, take a starter in anything that clears the filter above 75, cap me at $200 per name and $600 for the day." It's a hosted quant desk, not a chat window you have to babysit.
Risk rails & routing
- Fail-closed limits. Daily and per-trade USD caps, an allowlist with cooldowns, and a one-tap kill switch. If a trade can't be priced or a limit is hit, the agent refuses — the safe default is not to trade.
- MEV-protected routing. Orders go out through private relays by default, so they don't sit in the public mempool to be sandwiched or front-run. You keep the price you were quoted.
- Full audit trail & PnL. Every action is logged with its reasoning, and profit and loss is attributed call by call — the same receipts that grade the harness.
Autonomy modes
- Advisor — the agent only suggests; every trade is yours to place.
- Confirm-to-fire — it prepares the trade and the receipts; one tap sends it.
- Auto within limits — hands-off inside guardrails you define (max size, allowed chains, daily caps). Pull it back any time.
Screened before it signs. Every route is checked for honeypots, malicious contracts, and prompt-injection before a transaction is ever prepared. Speed never comes at the cost of walking into a trap.
Wallet & custody
- A wallet per user, non-custodial. A wallet is spun up for you and the keys are yours. You can withdraw at any time.
- No deposit address we control. There is no managed account and no way for us to move your funds.
- Your autonomy dial. The agent only acts as far as you let it, and approving or authorizing a trade is always your action.
$SAI & access Pre-launch
Shibalytics AI ($SAI) is the key to the pack. It has one job — no transfer tax, no fee tricks. Holding it unlocks the agent at full strength.
| Tier | What you get |
|---|---|
| Free | A taste of the hunt — limited daily runs and the base agent. |
| Hold $SAI | Unlimited hunts, the full harness, deeper reasoning, priority execution and higher autonomy limits. |
Contract address. $SAI has not launched yet. The official contract address will be published here and on the homepage at launch — anything claiming to be $SAI before then is not us.
Multichain
The intelligence layer is chain-agnostic; only the data adapters differ. The rollout follows the liquidity:
| Chain | Why | State |
|---|---|---|
| Ethereum | Deepest wallet history and the richest deployer forensics — the best place to prove the engine. | First |
| Base | Where memecoin and AI-agent volume is migrating. | Next |
| Robinhood | New retail rails and a fresh launch surface to read early. | Planned |
| BNB | High-velocity launch culture with its own metas. | Planned |
Roadmap
| Phase | Scope | State |
|---|---|---|
| P0 | On-chain signal engine reading live mainnet state (proven against live tokens). | Live |
| P1 | The backtested filter, the on-chain + social + incentive fusion, and the hosted trading harness. | In progress |
| P2 | Plain-English execution, the non-custodial wallet, order types and the autonomy dial. | Next |
| P3 | Base, then Robinhood and BNB; the full transparency dashboard; $SAI launch. | Planned |
FAQ
Do I need to set anything up?
No. Shibalytics is hosted — there are no API keys to wire and no model to choose. A non-custodial wallet is spun up for you, and you talk to the agent in plain English.
How is this different from a general "onchain agent"?
A chat bot with a wallet still needs you to know what to buy. Shibalytics brings the intelligence — the backtested filter, on-chain forensics, and the incentive radar — so it knows what to hunt before it ever asks you a question, and it grades every call afterwards.
Does a high score mean "buy"?
No. The score describes how strongly the evidence lines up; it never predicts price. Treat it as risk homework — the receipts are attached for exactly that.
Is the incentive radar accusing influencers of crimes?
No. It surfaces probabilistic signals from public data — timing, wallet linkage, coordination patterns — as inputs to a trading decision, not accusations, and reads public posts, not private group internals.
Who controls my funds?
You do. The wallet is non-custodial — the keys are yours, you can withdraw anytime, and the agent only trades as far as your autonomy setting allows.
Which chains, and when?
Ethereum first, because it has the deepest history to prove the engine on. Base is next, then Robinhood and BNB.
Security & disclaimers
Shibalytics is an information and analytics tool that can execute transactions on your instruction. Nothing here is financial, investment, legal, or tax advice, and nothing here is a solicitation to buy or sell any asset.
- Not advice. Scores and verdicts describe signals and on-chain state; they do not predict price or guarantee safety. Always do your own research.
- Probabilistic, not accusatory. On-chain and incentive scores are inferences from public data — signals, not proof or allegations.
- Risk of total loss. Memecoins are extremely volatile and many go to zero. Never trade more than you can afford to lose completely.
- Non-custodial. Shibalytics never holds your keys or your funds. You are responsible for your own security and for every trade you authorize.
- Best-effort data & screening. On-chain and social data can be manipulated and tooling can have bugs. Contract screening reduces risk; it does not remove it.