Market signal engine

AI system · Market intelligence

This system reads crypto data before anyone else does.

Before

Six paid terminals, opened one by one, every single morning
Hours of reading before a single decision got made
No record of why past calls failed, so the same setups repeated
Headlines with no direction, no weight, no context

After

13 collectors feeding one briefing, delivered on schedule
Every signal logged the second it fires, graded later, never rewritten
An 8-page desk replacing the terminal-by-terminal routine
News, macro, on-chain & derivatives read and rated automatically

Overview

A private investor running significant crypto capital wanted what institutions have: a proprietary system tracking his positions, his portfolio, and the market beneath both.

Crypto is unusually generous with data. Exchange flows, funding rates, open interest, liquidation maps, whale wallets, ETF movements: nearly all of it is public and available through APIs, which is not true of equities. That openness makes the market more efficient in theory. In practice it does the opposite: there is far more public data than any person can read, so the advantage goes to whoever processes it fastest. A trader working by hand is competing against desks that have used AI for years.

SetMyClaw built a proprietary intelligence system that collects from thirteen sources on automated cycles, runs a multi-agent pipeline over them, and delivers signals and portfolio state in one briefing. The edge isn't secret information. It's reading public information faster than the people on the other side of the trade.

Challenges

He had the capital, the conviction and the data subscriptions. What he didn't have was a system that could read faster than a person and keep an honest score.

Six-plus paid terminals checked by hand before any decision got made
No permanent record of past calls, so losing setups were free to repeat
News arriving as raw headlines with no direction or weight attached
Macro events sitting in a separate calendar, easy to miss
Off-the-shelf bots that execute a trade but can never explain the reasoning
Several requested metrics not actually available on the APIs he was paying for
In-house engineers who could execute defined tasks but couldn't architect a system at this level

What we built

Phase 1: The data layer

Thirteen collectors on automated cycles, all writing to one place so every strategy reads from the same source of truth.

Binance: price, funding, open interest, long/short ratios, order book, klines
Hyperliquid: funding, OI, depth, liquidation zones from 3x to 50x, whale positions
CoinGlass: cross-exchange funding, OI, liquidations, ETF flows, CVD
CryptoQuant: 33 endpoints including MVRV, SOPR, NUPL, flows, miner and whale data
Nansen smart-money leaderboard, LunarCrush social scores across 20 assets
FRED macro: CPI, Fed Funds, M2, DXY, S&P, Nasdaq, gold
News from six RSS sources with AI deep-dives on the top articles daily
Collectors are deterministic code, not AI, so the gathering step can't hallucinate

We audited every endpoint before writing a line of collection code. Every metric on the client's request list was tested against his own API keys and reported back with an honest verdict of what his subscriptions could and couldn't deliver. Where coverage fell short, we sourced better alternatives. Where a paid upgrade wouldn't have delivered what it promised, we said so and told him not to buy it.

Phase 2: The AI pipeline

A multi-agent briefing chain, each agent reading its own slice before a chief agent synthesises
AI-generated support and resistance levels, refreshed daily
A watcher monitoring live levels against stops and targets
An automatic scorer grading trade outcomes as they close
A memory layer holding signals, scorecard, patterns and journal
A RAG learning system built on ChromaDB
Split by source rather than by asset, so collection stays efficient and agents scale independently

The honesty layer. Every signal is logged the moment it fires, at the price it saw, and graded later against what actually happened. Nothing gets edited or quietly dropped. The system is built so it can't flatter itself.

Phase 3: The dashboard

An eight-page dashboard, mobile responsive, behind Cloudflare Access.

Overview: price ticker, Fear & Greed gauge, macro strip, live alerts
Charts: TradingView with toggleable overlays for MA, EMA, Bollinger, S/R and liquidation zones
Signals: active watchlist and market signal cards
On-chain, Social, News, Assets and Briefing pages, each scoped to the asset in view
Press anything and it explains itself: what it is, where the number comes from, how fresh it is

Phase 4: Delivery and control

Briefings to Telegram DM and channel, plus HTML email with chart snapshots
Per-alert routing: what reaches the DM, what reaches the group, what goes to the daily digest
A schedules panel showing every recurring job in one editable place
Two separate chats: an analysis chat that is read-only by construction and physically cannot break the system, and a building chat that goes through review before anything lands
Alert enforcement rebuilt so the trader's own settings are actually obeyed across every path

Phase 5: The strategy lab

Create, activate, pause and delete strategies directly
Rolling-forward testing: tune on one stretch of history, prove on the next, the way a trader would have actually traded it
Per-strategy record showing win rate, fire rate, and how many variations were tried before a candidate survived
New patterns noticed live, but only trusted after passing the same test against history

Results

Live scorecard from the system: 5 of 5 closed trades won since launch.

13 sources → 1 briefing

6 terminals collapsed into a single morning read

0 repeat mistakes

Every failed thesis diagnosed, stored, and checked against future signals

Results shown are from one client's live system over a short period and a small number of trades. They are not typical, not guaranteed, and not indicative of future performance. SetMyClaw builds software; nothing on this page is trading or investment advice.

How It Works

Collectors pull the data

Thirteen sources on automated cycles, every 30 minutes, into one shared data layer.

The audit came first

Every requested metric was tested against the live APIs before build, so nothing was promised that the data couldn't support.

Specialist agents read their slice

Flow, chief and signal agents each handle their own domain before anything is synthesised.

The briefing gets assembled

One read combining on-chain, derivatives, social, news and macro, cross-checked against the memory layer.

Signals fire and get logged

Entry, stop and target, written to an immutable record at the moment of firing, with the price the system saw.

Outcomes get graded

The scorer grades every closed position automatically. The record is never edited to look better than it was.

He decides, the engine doesn't

No auto-execution, by design. The system finds and scores the setup; the trade is always his call.

Want an intelligence engine like this behind your decisions?Book a call

Ready to build your advantage

Let's identify the highest leverage opportunities in your business.

Book a strategy call and I'll show you exactly where AI can make the biggest impact.

Book a Strategy Call

✓Custom audit of your funnel
✓Prioritized automation opportunities
✓Clear next steps