AI-Native Trading Platform · UK Operator Today · Beta Tenants Next · Regulated Marketplace Ahead
Trading agents that watch themselves trade —
and get better.
The doctrine: long-only, sector-aware, multi-TF self-stopping. Every decision auditable. The 6-layer funnel below decides what to trade and when to stop.
SentinelHub is a multi-agent AI trading platform that brings institutional-grade discipline to UK retail traders running US ETF strategies on Alpaca. Nine Gemini agents — orchestrated through Google's Agent Development Kit and instrumented end-to-end with Arize Phoenix — research markets, validate risk through a six-layer doctrine, execute small bracket-protected trades, and reflect on their own traces to propose evidence-backed parameter refinements. The edge is not a smarter model: it is statistical persistence across many small wins, with every decision auditable to the span that produced it.
Avg self-directed retail return vs S&P 500
Illustrative · cumulative gap under benchmark
The problem
500M retail traders.
Most lose money.
Discipline doesn't scale.
Nine specialist agents
Analyse · Validate · Execute · Reflect
Already running
Equity
—
Alpaca paper
Open positions
—
live · broker truth
Decisions logged
—
MongoDB Atlas
Tools traced
—
Phoenix OTel
Phase roadmap
01
Validation
Today
02
Social paper trading
Q3 2026
03
Live execution
2027+
Trading agents that
watch themselves trade
— and get better.
SentinelHub
The Problem
Retail traders lose money systematically
- →Trade emotionally — buy high on FOMO, sell low on panic.
- →Ignore risk — no stop-losses, no position sizing, no drawdown limits.
- →Can't access institutional tooling — quant tech costs millions.
- →Copy people, not strategies — eToro lets you mirror a personality with no logic visibility, no audit, no risk controls.
The Solution
A platform where AI does the discipline
- ✓Auditable AI strategies — every decision logged with confidence score, factor breakdown, and skip reason.
- ✓Institutional risk by default — bracket orders, heat caps, drawdown circuit breakers from the first trade.
- ✓Self-improving agents — they query their own Phoenix traces and propose parameter refinements.
- ✓Copy strategies, not personalities — verifiable rules, transparent risk parameters, statistical track record.
The trading philosophy
Many small wins, never one big bet
Every other "AI trading" demo promises asymmetric upside on a few hero trades. SentinelHub does the opposite — the edge is statistical persistence across hundreds of small trades, not luck on any single one.
Target trade size
$20 – $100
Per closed trade. Sized so the platform can carry many bets without any single one moving the equity needle. The math relies on win-rate persistence across 100+ trades, not on being right today.
Risk per trade
≤ 1% equity
Bracket orders submitted atomically with every entry — entry + stop-loss + take-profit, never a naked position. Heat-cap, drawdown breakers, and correlation guards layered on top.
Survival before returns
Process > Outcome
We accept many small losses to capture the statistical edge. Capital preservation outranks any single trade. If a strategy needs convincing to fire, the confidence score is too low — skip it.
How it works
The hourly intelligence loop
Six times per US trading session, the daemon runs a five-stage cycle. Every stage is a separate agent with bounded tools, traced end-to-end in Arize Phoenix, with persisted memory in MongoDB Atlas.
Research
Market analyst pulls bars, indicators, and regime signals across the 100-ETF universe.
Assess
Macro sentinel scores Fed proximity, VIX regime, sector dislocations, news sentiment.
Risk Check
Risk manager validates position sizing, heat, drawdown breakers, correlation guards.
Execute
Trade executor places bracket orders via Alpaca: entry + stop-loss + take-profit, atomically.
Reflect
Reflection agent queries traces and outcomes, proposes parameter refinements with evidence.
Multi-agent orchestration
Nine specialists, two model tiers
Reasoning-heavy agents run on gemini-flash-latest (FAST tier) for synthesis and judgement. High-volume, narrowly-scoped agents run on gemini-3.1-flash-lite-preview(LITE tier) for cost-efficient deterministic work. Routing is the orchestrator's job.
| Agent | Role | Model Tier | Tools |
|---|---|---|---|
| sentinel_hub | Orchestrator — routes work to specialists | FAST | — |
| — | — | — | — |
| — | — | — | — |
| — | — | — | — |
| — | — | — | — |
| — | — | — | — |
| — | — | — | — |
| — | — | — | — |
| — | — | — | — |
MCP integration foundations · Six partner platforms wired into the agent tree
Built on six MCP-integrated platforms
Each partner solves a distinct problem in the SentinelHub stack. We don't reinvent — we compose. The differentiation is the orchestration that connects them, not any single integration in isolation. Every integration is on the product roadmap regardless of any other context.
Google Cloud · Gemini
FoundationHow we use it
All nine specialist agents run on Gemini through the Agent Development Kit (ADK). Reasoning-heavy work uses gemini-flash-latest (FAST tier); high-volume narrowly-scoped work uses gemini-3.1-flash-lite-preview (LITE tier).
Why valuable
Tiered model strategy gives us institutional reasoning at consumer-tier latency and cost. ADK provides the orchestration foundations — tool routing, function calling, structured output — without building it from scratch.
— agents · — tools · ~42 LLM calls/day · all traced
Arize Phoenix
CoreHow we use it
OpenInference auto-instrumentation captures every agent call, tool invocation, and Gemini token. Five native Phoenix widgets embedded directly in our /monitoring page. The reflection_agent queries its own traces to detect drift and propose parameter refinements — the self-improvement flywheel.
Why valuable
Every decision becomes auditable, queryable, and replayable. Without Phoenix the agents are opaque; with it they are introspectable — the prerequisite for getting smarter over time.
OTel live · 5 monitoring widgets · evaluator harness designed
MongoDB Atlas
CoreHow we use it
Persistent agent memory across decisions, snapshots, outcomes, sentiment, briefings, and strategies (live corpus + collection counts on the MongoDB integration page). Powers historical context queries ("what happened the last 5 times in this regime?") and aggregation pipelines (win-rate by sector × regime in real time).
Why valuable
Without persistent memory, agents are amnesiac — every session restarts cold. Atlas turns ephemeral reasoning into a compounding knowledge asset that gets richer with every trade.
Live document corpus growing daily — see /integrations/mongodb
Elastic
CoreHow we use it
Decision corpus search and pattern discovery across the audit trail. Elastic Cloud v9.5.0 cluster reachable; the elastic_analyst agent has 4 dedicated tools (index_trade_decision, search_trade_patterns, get_strategy_analytics, detect_decision_anomalies). A 3-tier fallback (Elastic BM25 → MongoDB Atlas Search → regex) means /decisions never returns empty during the rolling backfill window.
Why valuable
Once we cross 50K decisions, full-text and semantic search becomes the only way to mine for patterns: "show me decisions skipped because of FOMC proximity that would have been winners". A different shape of query than aggregation.
Cluster v9.5.0 live · 3-tier search fallback live · ELSER + dense_vector on Phase 2.5 backlog
Fivetran
EmergingHow we use it
Managed data pipeline from MongoDB Atlas → BigQuery warehouse for backtest reproducibility and multi-source analytics. Real Fivetran REST API path implemented; data_pipeline_manager agent surfaces connector inventory and freshness. BigQuery becomes the deterministic source of truth for backtests.
Why valuable
Backtests must be reproducible. Decisions must outlive the operational store. Fivetran turns "snapshot the operational DB on Tuesday" into a managed, monitored, audited pipeline — exactly what regulators will want when we go live.
REST API path live · operator registered · connector activation pending
Dynatrace
InfrastructureHow we use it
Full-stack infrastructure observability and the pre-trade infrastructure gate. Before every 30-minute scan, the risk_manager agent queries Dynatrace for active problems and anomalies. If infrastructure is degraded, signal confidence is halved; if critical, all trades are halted. BizEvents push every trading signal and order into Dynatrace Grail for cross-correlation with infrastructure metrics — answering 'did signals generated during infra degradation perform worse?'
Why valuable
Don't trade on signals generated by a degraded system. Dynatrace closes the loop between infrastructure reliability and trading confidence — a concept unique to production trading systems at scale. Infrastructure-aware trading.
Health gate · anomalies · BizEvent ingest — see /integrations/dynatrace
Other building blocks
The rest of the stack
Outside the sponsor program, these third-party capabilities round out the platform — broker execution, macro data, sentiment ingestion, and the multi-tenant infrastructure scheduled for Phase 2.
Alpaca Markets
LivePaper trading execution — bracket orders with broker-side stops
yfinance + Alpaca VIXY
LiveMacro signal: VIXY proxy ETF for VIX, treasury yield curve via yfinance (5Y/10Y/30Y). Fallback when FRED is unreachable.
FRED API
LiveAuthoritative US macro — CPI, jobs, inflation, VIX — fetched on release and cached.
X / Reddit / FED RSS
LiveThree-tier sentiment summarisation engine — Gemini Lite distils raw posts into market weather
Firebase Authentication
LiveIdentity + ID-token verification on every API call; multi-tenant custom claims land in Phase 2
Stripe Billing
PlannedSaaS subscription + marketplace commission collection (Phase 2)
Phase roadmap
From validation today to live execution at scale
A deliberately staged product: prove the engine first, then scale safely on paper trading where mistakes cost nothing, then graduate to live capital with regulatory backing. The colour of each phase reflects its current state.
Foundation & Validation
A working trading engine: — Gemini agents, — tools, a full audit trail, and live paper trading on Alpaca.
Capabilities
- ·Specialist agents orchestrated by gemini-flash-latest
- ·100-ETF universe across 14 asset-class buckets
- ·6-layer institutional risk stack (bracket orders, heat caps, drawdown breakers)
- ·Persisted decision corpus in MongoDB Atlas
- ·Real-time agent traces in Arize Phoenix (OTel auto-instrumentation)
- ·LLM-as-judge: Gemini scores every decision against a 6-criterion rubric (F1 LIVE)
- ·Self-improvement engine: cohort attribution + Gemini narration, live.
- ·Per-strategy recommendation panel — surfacing now.
- ·Closed-loop flywheel: annotate → judge → reflect → propose → operator-approve → apply. Foundation and persistence are live; the approval UI is rolling out.
- ·Full Next.js 16 dashboard — Market, Strategies, Portfolio, History, Insights, Flywheel, Monitoring, Data-Pipeline, Settings
Social Paper Trading
Open the platform to thousands of users on Firebase. Paper trading only — zero capital risk, but every strategy is a real, auditable AI. Users follow strategies (not personalities) and earn rewards when their followed strategy wins.
Capabilities
- ·Firebase Authentication + multi-tenant MongoDB
- ·Strategy Marketplace — publish, discover, copy strategies with full audit trail
- ·Follow & Earn — strategy creators earn commission from follower performance
- ·Conversational strategy creation: chat with AI Strategy Advisor to design custom rules
- ·Backtest-as-a-service before any strategy goes live
- ·Mobile-first responsive design + push notifications
- ·Stripe billing — Free / Pro / Enterprise tiers
Live Execution & Advanced AI
Real capital, real execution. Brokerage-integrated live trading with FCA-compliant onboarding. Self-improving AI proposes parameter optimisations from its own observability data — the system gets smarter every trade.
Capabilities
- ·Live brokerage execution (Alpaca live + Interactive Brokers)
- ·FCA-regulated onboarding for UK retail traders
- ·Advanced visualisation: 3D regime maps, factor attribution, scenario analysis
- ·Institutional features: SAML SSO, dedicated infrastructure, compliance reporting
- ·Multi-strategy ensemble: 6+ live strategies × 3 variants each (~20 in flight)
- ·Real-money paper-to-live promotion ladder with 60-trade evidence gate
Monetisation roadmap
From single-operator validation to a follower marketplace
We earn nothing during validation. The product becomes commercial in Phase 2, when paper-trading users follow strategies and creators earn commission on follower performance — a $10B+ copy-trading market with a unique twist: full algorithmic transparency.
Phase 1 — Now
£0
Operator-only validation phase. The engine proves the thesis on real bracket orders against Alpaca paper. No external revenue. No external users.
Phase 2 — Marketplace
SaaS + Commission
Free / Pro / Enterprise tiers. Strategy creators earn commission when followers' paper portfolios outperform benchmarks. Stripe billing.
Phase 3 — Live
Performance Fees
Live capital with FCA-regulated onboarding. AUM-based pricing for retail; enterprise licensing for funds and prop desks.
Platform credits
Built on a six-partner integration foundations
SentinelHub composes against the open-source and partner platforms above. Each integration is permanently on the product roadmap; the agent tree degrades gracefully when any single platform is unreachable.
The APP-051 doctrine
The six-layer funnel — what to trade, and when to stop
Long-only, sector-aware, multi-TF self-stopping. Each page in this dashboard surfaces a chip mapping it to one of these layers. The funnel runs top to bottom every day, then reflects and rinses-and-repeats.
- L0 — Macro tailwind
The global regime gate
A healthy US + global economic regime is the foundational gate at the top of the funnel. When the macro tailwind is green it enables long-only deployments; when it turns red the funnel skips the costly dislocation and sector steps entirely.
- L1 — Dislocation scan
Exogenous-event dislocation scan
News, sentiment, and political-event signals surface sectors that have been beaten down despite the macro tailwind — the unloved, oversold pockets a long-only operator can actually buy. Each candidate carries an evidence chain back to the event that caused it.
- L2 — Sector identification
From 13,000 tickers to a sector universe
The dislocated segment is filtered from the ~13,000 Alpaca-tradeable tickers down to a curated sector universe of ~5-12 ETFs (energy, materials, miners…), checked for liquidity and dispersion.
- L3 — Strategy activation
Bind a strategy template to a sector universe
A strategy template (RSI-2, Momentum, Squeeze, or custom) is bound to the sector universe to produce an active strategy deployment with its own parameters. Different sectors get different templates. The entry signal (L3a) inside a deployment always fires as a bracket order — a stop-loss and take-profit from the moment of submission, no naked entries.
- L4 — Multi-TF overheat
Self-stopping when the sector runs hot
A 30m / 1H / 3H / 1D confluence scorer watches each deployment's sector universe. When the sector overheats the deployment auto-pauses — no new entries — while existing bracket orders manage open positions to their natural exit. The platform never chases a move that has already mean-reverted.
- L5 — Reflection
Reflect, then rinse-and-repeat
At end of day the reflection engine partitions closed trades by strategy × sector × regime, attributes which factors drove the outcomes, and Gemini narrates the cohort evidence into an operator-approvable proposal for the next deployment's parameters. The next day re-runs Layers 0-2 to spot the next dislocation.
Phase 1 — Foundation & Validation
The engine ships. External validation is next.
Every page in this dashboard renders live data from a real trading daemon placing real bracket-protected orders on Alpaca paper. We are entering the design-partner beta phase: a small cohort of operators mirror strategies on their own paper accounts and stress-test the multi-tenant surface before live capital is ever on the table. Strategies first, capital later.