Aelix Propel · WealthTech and investment AI
Run more portfolios. Personalize every one. Prove compliance on every trade.
Aelix Propel gives advisory firms an institutional-grade operating layer for managed portfolios. Automated rebalancing, continuous tax-loss harvesting, ESG-aware direct indexing, a pre-trade mandate engine, and a quantitative risk engine. So a single advisor can serve far more households without watering down personalization or weakening controls.
Aelix Agents · Agentic AI · Now in active build
An agentic layer over the portfolio spine.
Propel now includes an agentic layer. Autonomous, tool-using agents that act through Propel's governed APIs, do the repetitive portfolio work, and escalate the judgment calls to a human. Every agent is human-in-the-loop by default, runs inside your perimeter, and writes each step to the same append-only audit trail as the rest of the platform.
Portfolio Drift & TLH Agent.
Scans for drift and tax-loss-harvesting opportunities and prepares mandate-checked rebalancing proposals for the advisor to sign off.
Proposal Drafting Agent.
Assembles a personalized, compliance-checked client proposal from the model portfolio and risk profile.
Client Outreach Agent.
Drafts proactive, suitability-aware outreach when a portfolio or life event warrants it.
Agents act only through permissioned, least-privilege APIs, and every proposed order still passes the pre-trade mandate engine. They never execute a trade without advisor approval. The reasoning model is a swappable component: run a local model on-prem for full data residency, or a hosted model where policy allows.
Now in active build. The first agents ship as configurable, versioned workflows.
The problem
Advisory teams are squeezed from both ends.
Advisory teams are squeezed from both ends. Clients expect personalized, tax-aware, values-aligned portfolios that historically only the largest accounts could justify. Regulators expect documented evidence that every order respected the client's mandate and the firm's risk limits. And the math of the business demands that each advisor manage more households, not fewer.
You cannot solve that with more spreadsheets and a quarterly rebalance. Manual drift checks miss windows. Tax-loss opportunities expire unnoticed. ESG exclusions get applied inconsistently. And when an examiner asks why a particular trade went through, the answer lives in someone's memory instead of an audit trail.
Aelix Propel closes that gap. It automates the repetitive portfolio work, enforces every client mandate and regulatory limit before an order can leave the desk, and records what happened in an append-only audit log. So advisors spend their time on relationships and judgment, not reconciliation.
Available today
Shipped, running code. Sandbox-by-default, live by configuration.
Every capability below is implemented in the product today against a deterministic sandbox for execution and market data. Brokerage and live-feed connectors are credential-pluggable and ship as the default sandbox until onboarding (see On the roadmap).
Drift-based smart rebalancing
Propel continuously compares each account against its assigned model portfolio, computes per-sleeve drift, and generates a rebalance proposal the moment drift crosses the model's threshold. Each proposal arrives as a concrete set of buy/sell legs with target versus current weights. Ready for advisor review, not a vague 'you are off target' flag. Drift work that used to be a quarterly chore becomes an always-on watch on every household.
Continuous tax-loss harvesting
Propel scans open tax lots for unrealized losses that clear configurable dollar and percentage thresholds, then pairs each loss-realizing sale with a substantially-similar replacement (for example VTI into ITOT, VOO into IVV) so the client keeps market exposure while booking the loss. Lots inside a 30-day wash-sale window are automatically skipped. Each opportunity surfaces as a high-urgency alert for the advisor with the loss amount, the lot, and the suggested swap. So tax alpha is captured in the moment instead of discovered at year-end.
ESG-aware direct indexing
Run a direct-indexed sleeve that holds the underlying constituents of an index while stripping out the companies a specific client objects to. Per-client exclusions (by symbol, by sector, or by minimum ESG score) live as mandates and are honored automatically at rebalance time. Each client gets an index built around their values, at a scale that simply was not viable when exclusions were managed by hand.
Pre-trade mandate engine
This is the control that lets you say yes to more customization without losing sleep. Every child order is evaluated against the client's active mandates before it can route. Sector caps, single-position concentration limits, restricted lists, asset-class floors, ESG exclusions, and SEC 15c3-5 notional checks. Violating orders are intercepted and held back (critical breaches blocked outright, softer ones flagged for review) while clean orders flow through. Every interception is written to an append-only, exportable audit trail with the rule that fired, the severity, and the reason. The evidence a compliance officer wants when the question comes.
Quantitative risk and stress-testing engine
A new quantitative risk layer turns 'are we exposed?' into numbers you can act on. Propel computes parametric and historical Value-at-Risk and CVaR over a portfolio's return series, and runs scenario stress tests against canonical shock vectors. The 2008 global financial crisis, the 2020 COVID drawdown, the 2022 rate shock, and a generic minus-two-sigma day. Reporting implied P&L and each asset class's contribution. Alongside it, a trained gradient-boosted classifier produces a market-shock probability from engineered volatility, drawdown, tail, and macro features. This is defensible early-warning analytics: scenario stress testing plus a calibrated shock signal. Not a guarantee or a prediction.
Real ML security scoring
Propel's scoring runs on genuinely trained models, not random placeholders. A gradient-boosted scorer rates securities and surfaces rebalance, ESG, and tax-harvest signals, each with importance-weighted driver explanations so an advisor sees why a score moved, not just the number. If a model artifact is ever unavailable, the system degrades gracefully to a transparent rule-based fallback. It never goes dark and never fabricates a result.
AI proposals, dual portals, and enterprise controls
Generate client-ready PDF proposals on demand. Give advisors a workspace for models, drift, mandates, and the execution blotter, and give clients their own portal for statements, holdings, and tax estimates. Underneath sits role-based access control, multi-factor authentication, and a hash-chained, append-only audit log across every sensitive action.
How it works
From model definition to settled execution. One spine.
01
Define the model.
Assign each account a model portfolio and layer on the client's mandates. Sector caps, concentration limits, restricted lists, ESG exclusions, and regulatory notional checks.
02
Detect and propose.
Propel watches for drift and tax-loss opportunities continuously, generating rebalance proposals and TLH alerts with concrete, reviewable legs.
03
Check every order pre-trade.
Approved orders are aggregated into block trades and split into per-account child orders. The mandate engine evaluates each child before routing. Clean orders proceed. Violations are intercepted and logged.
04
Aggregate and execute.
Cleared children route through the Smart Order Router as an aggregated block, fills are allocated back to each account, and positions and tax lots update automatically. Every step captured in the audit trail.
The honest proof
Numbers we can defend, framed with their conditions.
4 trained ML models in production paths.
Market-shock, ESG, drift, and tax-harvest. With reported test ROC-AUC of 0.97 to 0.99 on their validation sets.
4 canonical stress scenarios.
Plus parametric and historical VaR/CVaR, computable on any portfolio.
6+ mandate rule types enforced pre-trade.
Including SEC 15c3-5 notional checks.
100% of intercepted orders.
Written to an append-only, exportable audit trail.
Graceful degradation everywhere.
If a model artifact is missing, scoring falls back to transparent rules rather than failing.
A note on honesty
Propel's ML models are trained on synthetic data generated with documented domain structure, not on real market history or client portfolios yet. The metrics above reflect validation against that simulator. We treat the scores as well-calibrated ordinal signals for advisor decision support. And we will say so until the models are retrained on licensed historical data. We publish no AUM or client-count figures we cannot substantiate.
On the roadmap
Configuration and onboarding, not rewrites. The interfaces are already in place.
The architecture is built so these become configuration and onboarding, not rewrites. The interfaces are already in place.
Live brokerage execution (Schwab FIX 4.4).
The broker adapter is credential-pluggable today. Selecting the live profile validates credentials and is wired to a real FIX session. It intentionally does not fabricate FIX traffic. Live order execution becomes available on completion of Schwab institutional FIX onboarding.
Live market data.
Real market-data adapters (IEX / Polygon-style) are scaffolded and gated on an API key, with the deterministic sandbox feed as the default until a data subscription is in place.
Models trained on real historical data.
Today's models train on realistic synthetic series with the right directional structure. The feature builders stay the same. Retraining on licensed historical and portfolio data is the next step.
Custodian selection.
The adapter pattern already supports multiple custodians (Apex / Pershing / Schwab / Fidelity / IBKR). The choice is a commercial decision.
Workforce SSO.
Single sign-on (Okta / Azure AD / Auth0) for advisor login.
SOC 2 Type II / ISO 27001.
The control posture exists today. Formal certification is on the roadmap.
Security and compliance
Engineered for the controls advisory firms and their examiners expect.
Role-based access control and multi-factor authentication.
On every privileged action.
Append-only, hash-chained audit log.
Capturing mandate interceptions, trades, configuration changes, and proposal generation. Exportable for examination.
Pre-trade enforcement.
Of client mandates and SEC 15c3-5 notional limits, so controls are preventive, not just detective.
GDPR-ready by design.
With privacy and consent-oriented controls.
SOC 2-aligned posture.
The technical controls of a SOC 2 program are implemented today. Formal SOC 2 Type II and ISO 27001 certification are on the roadmap.
See it for yourself
Ready to give every advisor institutional reach?
See drift detection, the mandate engine, the risk engine, and aggregated execution running against live sandbox data. Then map the path to live brokerage and market feeds for your firm.
Frequently asked
Questions teams ask in the second meeting.
Does Propel trade in my clients' accounts automatically?
Not autonomously. Propel is a hybrid model: the system proposes (rebalances, TLH opportunities, security scores) and a credentialed advisor reviews and approves before anything routes. Live order execution is on the roadmap pending Schwab FIX onboarding. Today execution runs against a deterministic sandbox.
How does the mandate engine prevent a non-compliant trade?
Every order is checked before it can route. The engine evaluates the client's active mandates. Sector caps, concentration limits, restricted lists, asset-class floors, ESG exclusions, and SEC 15c3-5 notional checks. Critical violations are blocked, softer ones are flagged for review, and every interception is written to an exportable audit trail with the rule, severity, and reason.
Is the 'market shock prediction' a real forecast?
We present it honestly as what it is: scenario stress testing against canonical historical shocks (2008, 2020, 2022, minus-two-sigma) plus parametric and historical VaR/CVaR, and a trained shock-probability signal from volatility, drawdown, tail, and macro features. It is defensible early-warning analytics for decision support. Not a guarantee, not a price forecast.
What are the ML models actually trained on?
Today they are trained on synthetic data generated with documented, defensible domain structure. Not real market history or client portfolios. The models are genuinely predictive and ship with driver explanations and a rule-based fallback. Retraining on licensed historical data is the next step, and we will not claim otherwise until it is done.
How does tax-loss harvesting avoid wash sales?
The scanner skips any lot with a purchase of the same security in the trailing 30-day window and pairs each loss-realizing sale with a substantially-similar replacement, so the client keeps market exposure while booking the loss. Account-level wash-sale handling ships today, with household-level checks on the roadmap.
Need this tailored to your environment?
Every Aelix product can be configured, extended, or built bespoke for your industry, data sources, and compliance constraints. Talk to our engineers about what would change.
Configurable workflows
Adapt rules, thresholds, and approval flows to match your operational policies.
Custom data integrations
Connect to your specific ERP, MES, SCADA, CRM, or proprietary systems.
Bespoke modules
Build product extensions tailored to your industry, region, or compliance needs.