Alepes helps you define how new money should flow into your portfolio, then directs contributions toward the holdings that need them most.
Rules-based capital allocation, with every decision explained.
Preview only — live transfers and live brokerage execution are not currently available.
Payroll deposit detected
How it works
Your cash-flow policy decides whether and how much to deploy from new money — triggers, reserves, and caps. Your investment policy decides where it goes: the target allocation and bands that define your formation. Keeping them separate is what makes every decision explainable.
Cash-flow rules
WHEN
a qualifying deposit is detected
IF
checking remains above $5,000
THEN
invest 40% of the deployable amount
Formation
Formation is how close your portfolio is to its target allocation. Holdings drift over time; Alepes uses each new contribution to move the school back toward formation — underweight first, without forced selling.
| Symbol | Target | Current | Action |
|---|---|---|---|
| VTI | 50% | 44% | Buy |
| AAPL | 25% | 30% | Hold |
| VXUS | 25% | 26% | Hold |
New contribution → directed primarily toward VTI, the most underweight holding.
Shadow Mode
Shadow Mode evaluates your rules and shows exactly what would happen — a full plan, with nothing executed.
Explainability
Future capability
Designed for what comes next, Alepes may one day let models influence a Formation — at the level of a single holding, of the whole portfolio, or both. A model would never submit an order or move money. It produces investment intent, an input that Alepes resolves through your policy, constraints, and execution rules.
The model is an input to Alepes policy — not the authority. Every step after intent is Alepes's own deterministic machinery, unchanged.
A model assigned to one holding evaluates that asset and produces a signal or a constrained intent — a valuation read, a momentum view, a quality flag, a volatility measure, or a factor score. It never produces a broker order directly.
A Formation-wide model evaluates the portfolio as a whole and may propose target weights or portfolio-level adjustments. Alepes still validates the resulting Formation against your rules and safety constraints.
Formation model portfolio-risk-v4 AAPL → valuation-quality-v2 MSFT → momentum-v1 VTI → (no holding model)
When both levels exist, composition is explicit: holding models contribute asset-specific signals while a Formation-level model resolves those signals into a coherent portfolio view. There is no “most specific model wins.” Any override behavior is a policy you configure, not a default.
Example
Model: AAPL target = 40%
Constraint: max single holding = 20%
Result: Alepes caps the target at 20%
Example
Model: allocate $800
Capital plan: only $240 investable
Result: Alepes can allocate at most $240
A future model would first run in Shadow Mode, where Alepes compares the current Formation, your target Formation, and a model-resolved Formation — side by side, with nothing executed. This is a simulation to review, not a live allocation.
Current Formation
40 / 35 / 25
User target
50 / 30 / 20
Model-resolved target
44 / 38 / 18
A new flow of $500 would be allocated against the resolved target, and the difference would be explained — not silently adopted.
A future Alepes decision could record: which model was used, its version, the input snapshot, the resulting intent, the policy constraints applied, the final resolved Formation, the Shadow or execution decision, and an explanation of why the result changed. Alepes should be able to explain not only what a model suggested, but how Alepes transformed that suggestion into an allowed decision.
Alepes owns the model system. External libraries may provide model implementations.
Models produce investment intent. Alepes resolves that intent into a Formation and applies its existing deterministic financial and safety policies.
Momentum, valuation, volatility targeting, contribution tilt, and factor scoring — simple, explainable, and policy-bounded.
Proven portfolio-optimization libraries may one day provide Formation-wide models such as minimum volatility, maximum Sharpe, Black-Litterman, or hierarchical risk parity. This is an example of a type of integration, not something integrated today.
Alepes is designed to host multiple model types without making any one quantitative framework the financial brain of the product.
Models produce opinions. Alepes decides what is allowed.
Product preview
Static mock views of the product. No live connections, no real accounts.
| Symbol | Target | Current | Action |
|---|---|---|---|
| VTI | 50% | 44% | Buy |
| AAPL | 25% | 30% | Hold |
| VXUS | 25% | 26% | Hold |
New contribution → directed primarily toward VTI, the most underweight holding.
Safety guardrails
No order or transfer is submitted unless a separately authorized execution capability exists and the user has explicitly granted it. Current state: preview only.
Reserves, caps, and monthly maximums are user-authored. Nothing deploys beyond what the user's rules permit.
Every decision is traceable. Immutable evidence preserves original provenance — corrections append history rather than rewrite it.
Provider tokens and secrets never appear in logs or repository files. Credentials remain protected.
Simulation parity means Shadow and future execution use the same deterministic plan — differing only at a separately controlled execution gate. Live transfers are not currently available.
Missing, stale, contradictory, or unauthorized inputs never become permission to invest or move money.
Alepes provides automation infrastructure, not investment advice. All investment decisions remain yours. This preview is for informational purposes only and is not a recommendation to buy or sell any security.
Release roadmap
Stages below are sequenced by evidence and dependency. Version labels are proposals subject to the repository's actual release conventions. Stages may be split, deferred, or rejected following risk review. No stage implicitly authorizes implementation of the next.
Real posted deposit provenance, deterministic recomputation, persisted audit/outbox, zero external mutations, credentialed independent certification. PR #6 open; live deposit proof absent.
Required acceptance evidence
Real deposit event with redacted provenance yielding identical plan under independent recomputation
Authenticated account ownership, permissioned connections, persistent rules, end-to-end explainability, revocable approval queue with no order path.
Required acceptance evidence
Authenticated user completes full authorized read-only journey; reproduced decisions from immutable inputs; revoked/expired consent fails closed
Separate operator approval, idempotent scoped capabilities, mock/sandbox replay, complete reconciliation state machine. Decision gate required before work.
Required acceptance evidence
Strong mock-provider conformance; provider sandbox/paper testing only if separately approved
External provider rights, regulatory review, explicit human consent, independent executed-order/settlement evidence, hard limits, kill switch. Separate approval only.
Required acceptance evidence
Legal analysis, approved broker/bank contracts, user consents, certified reconciliation, verified external settlement
Penetration review, privacy/retention, recovery drills, incident handling, onboarding accessibility, user experience data.
Required acceptance evidence
Documented and exercised drills; comprehensive logging without secrets; encrypted backups and restore drills
All relevant certification, regulatory, operational, adoption, and release gates accepted.
Required acceptance evidence
Explicit operational/regulatory sign-off; proven retention/value; measured cohort economics
Separate research program
Research programs explore possible investment opportunities. They must remain distinct from the production financial core and cannot submit orders or override a user's chosen portfolio. A future investment-recommendation product would require a separate charter, regulatory analysis, and explicit authorization.
Which momentum signals beat simple benchmarks after costs?
Required next evidence
Historical point-in-time replays, adverse regimes, held-out test, independent risk review
Unchanged rule
Preserve earlier negative/NO-GO capital-allocation findings until evidence changes
Are cross-security graph residuals useful beyond sector/linear baselines?
Required next evidence
Historical network construction, graph-stability & correlation-breakdown stress, leakage attacks
Unchanged rule
Synthetic performance is not live alpha
Can positions bought before public earnings create positive expected net value after release?
Required next evidence
Release-timestamp-correct data, historical bid/ask option chains, IV-crush, expiration, costs, held-out event studies
Unchanged rule
No uncovered shorts or fictitious option fills
Do financially sound businesses bought at a sufficient margin of safety outperform suitable long-term benchmarks?
Required next evidence
Point-in-time financial statements, honest owner-earnings/quality assumptions, value-trap & multi-year drawdown tests
Unchanged rule
Valuation scenarios are estimates, not promises
R0 data-license/architecture approval → R1 synthetic deterministic mathematical proof → R2 independent invariants and adversary → R3 point-in-time empirical data → R4 untouched out-of-sample comparison after realistic costs → R5 human decision on further research or separately scoped product/regulatory proposal. Negative and inconclusive results are legitimate endings.