AYVID — Asymmetric Yield Via Intelligent Decisions
AYVID · AI-Native Hedge Fund

The First Fully
Autonomous
AI Hedge Fund.

Asymmetric  Yield  Via  Intelligent  Decisions

Six asset classes. 300+ instruments. Zero human traders. LLM agent swarms read every 10-K, earnings call, and central-bank print, trace consequences through a live causal graph of the global economy, and route sized, hedged orders across equities, FX, commodities, crypto, ETFs, and bonds — end-to-end, fully autonomous.

See the engineSpeak with the desk
Live · IB Paper6 asset classes · fully wiredNeo4j · causal graph  ·  30 exchanges
Live engine · IB PaperCausal graph

Operations console

Trading the second-order
consequences, live.

Each panel below is wired to the same engine that places the trades. Walk-forward 2018 → 2025 across 20 named historical shocks. No hindsight. No staging data.

  • LangGraph swarm
  • Neo4j live graph
  • IB execution
  • C++ kill-switch
Causal supply-chain graphLangGraph agent swarmNeo4j live topologyEnd-to-end executionReal-time SEC + 10-K parseSub-100ms tier-1 hedge20 global indices · 24/5Walk-forward 2018→2025C++ kill-switch watchdog
Open to allocator capital

Hello — we are AYVID.

An AI-native hedge fund built end-to-end around causal reasoning. We don’t bolt LLMs onto a quant stack. We use them to find strategies that didn’t exist before.

About the firm

Capabilities

Causal AIAgent swarms6 asset classesNeo4j300+ instrumentsInteractive BrokersWalk-forward backtestsMulti-asset execution

Milestones

  • AYVID founded

    Apr 2026

    Engine v0 · modular monolith

  • Causal graph + agent swarm

    Apr 2026

    Neo4j + LangGraph integration

  • IB Paper · live shock catalog

    Apr 2026

    20 named shocks · walk-forward

  • YC S26 · seed conversations

    In progress

    Allocator outreach

Five-layer alpha architecture

From language to execution.

Five layers turn unstructured language into sized, risk-limited orders across every major asset class — fully autonomous.

See the architecture
01Layer 01

Supply-Chain Intelligence

A live Neo4j graph of 12,000+ supplier/customer relationships, dynamically enriched from SEC EDGAR filings. Graph centrality identifies the critical nodes that move markets.

02Layer 02

Shock Detection & Routing

NER pipelines ingest every 10-K, earnings call, and central-bank print in real time. Each shock is classified, geotagged, and routed across 6 asset classes — equities, FX, commodities, crypto, ETFs, and bonds.

03Layer 03

Multi-Agent Debate

Seven LLM specialists — Strategist, Mapper, Researchers, Analyst, Pattern, Auditor, Supervisor — debate every signal. Bayesian reconciliation across agents is how we estimate conviction. Disagreement itself is a signal.

04Layer 04

Premium Harvest

Signals route to 300+ instruments across 30 exchanges. Sized, hedged, risk-limited orders execute via Interactive Brokers in under 100 ms. Multi-asset cost model with realistic slippage.

05Layer 05

Kill Switch

An independent C++ watchdog runs in its own process. Position-level VaR, drawdown, and concentration limits enforced pre-trade. Can flatten the entire book in milliseconds.

Built on Anthropic

The buy-side production of Claude for Financial Services.

AYVID is the hedge-fund production deployment of Anthropic’s financial-services stack. We adopt the file-based Claude Skills convention, wire the buy-side slice of the upstream MCP connector registry directly into our continuous ingestion worker, and layer five proprietary skills on top — the parts of the AYVID stack (shock thesis, causal validation, Kelly sizing, post-mortem, bull/bear debate) that have no upstream analogue.

15 skills · 5 connectors
Adapted

Earnings analysis

from anthropic/financial-services · equity-research

Adapted

Earnings preview

consensus + IV-implied move + swing items

Adapted

Thesis tracker

nightly INTACT / WEAKENED / INVALIDATED verdicts

Adapted

Catalyst calendar

30-day forward, ranks priority dates

Adapted

Idea generation

ranked candidates from screener × shocks

Adapted

Model update

assumption delta + fair-value range

Adapted

Morning note

06:30 ET situational-awareness brief

Adapted

Sector overview

leaders, laggards, stress map

Adapted

DCF model

fundamental anchor on shock theses

Adapted

Comps analysis

peer multiples cross-check

AYVID-native

Bull / bear / facilitator debate

researcher LangGraph node

AYVID-native

Shock thesis

directional thesis anchored on causal chain

AYVID-native

Causal validation

edge adjudication against return data

AYVID-native

Kelly sizing

plain-English explanation of Half-Kelly

AYVID-native

Post-mortem

trade rationale + priors-update suggestion

MCP data connectors

  • MT Newswires

    Real-time news → shock detection feed

  • Aiera

    Earnings call transcripts

  • FactSet

    Multi-asset fundamentals + consensus

  • Moody's

    Credit ratings → ETFs & bonds sizing

  • S&P Global

    Capital IQ tear sheets

Each connector is opt-in via a single env var. When the URL is unset, AYVID falls back to its existing extractor stack (news scraper, OpenBB, Finnhub, SEC EDGAR) — no behavioural change.

What we run

Capabilities, not promises.

See it live

  • Causal supply-chain graph

    Live Neo4j topology of suppliers, customers, indices, and shared inputs. Shock propagation in milliseconds.

  • Agent-swarm research

    Seven LLM specialists debate each signal. Disagreement is itself a signal that the trade is fragile.

  • End-to-end execution

    From 8-K to broker order in one modular monolith. No humans on the trade path. Tier 1 sub-100ms.

  • Risk-first by design

    Independent C++ kill-switch watchdog. Position-level VaR, drawdown, and concentration limits enforced pre-trade.

Walk-forward backtestsLive shock catalog · 20 events6 asset classes · 300+ instrumentsMulti-asset cost modelReal-time PnL attributionOperations console

Why AYVID

What separates us from a traditional fund.

Most quants will bolt LLMs onto an existing book. We started from language and worked outward to execution. The architecture is the edge.

AYVID

Causal reasoning

We trace each shock through suppliers, customers, and indices. We trade the unobvious second-order beneficiaries.

Traditional fund

Factor regression

Lagged factors, momentum residuals, value premia. Linear models on non-linear consequences.

AYVID

Agent swarms

Seven specialists debate every signal. Disagreement is conviction-discounted.

Traditional fund

Single research analyst

One PM's mental model bottlenecks the entire firm. Disagreement is dampened, not measured.

AYVID

End-to-end automation

8-K to broker order in one system. No humans on the trade path. Risk pre-checked.

Traditional fund

Human in the loop

Analysts hand-off to traders. Latency in seconds-to-minutes. Behavioural drift in execution.

AYVID

Verifiable in operations

Live shock catalog, walk-forward fund performance, runner status — all reproducible from the console.

Traditional fund

Quarterly tear sheets

Aggregated, smoothed, lagged. Hard to attribute alpha to actual decisions.

By the numbers

All figures are verifiable in the console.

Verify live
  • 300+
    Instruments
    across 6 asset classes
  • 30
    Exchanges worldwide
    US, EU, Asia, EM
  • 6
    Asset classes live
    equities · FX · commodities · crypto · ETFs · bonds
  • 0
    Human traders
    fully autonomous

Asset classes we trade

One shock. Six expressions.

A single geopolitical event simultaneously routes to equities, FX, commodities, crypto, ETFs, and bonds. Multi-asset expression is how we capture the full consequence surface.

  • Equities

    ~50 names

    Shock-driven long/short, causal beneficiaries

  • Index F&O

    20 global indices

    Hedged macro, straddles, verticals

  • FX

    34 pairs

    Macro trend, carry trade, mean reversion

  • Commodities

    17 futures

    Supply shock, momentum, seasonal edges

  • Crypto

    22 assets

    Momentum, supply-chain expression

  • ETFs & Bonds

    37 ETFs + 4 rate futures

    Sector rotation, factor momentum, rate curve

Voices

The thesis, in other people’s words.

  • RFS · AI-native hedge funds

    “The next Renaissance, Bridgewater, and D.E. Shaw will be built on AI. The biggest funds will be slow to adapt.”

    CH

    Charlie Holtz

    Y Combinator · Request for Startups

  • RFS · Strategy generation

    “The alpha is in using AI to come up with strategies that didn’t exist before — language comprehension, causal reasoning across thousands of filings.”

    CH

    Charlie Holtz

    Y Combinator · Request for Startups

  • Internal · Causal edge

    “Causal inference on supply chains is the most underweighted edge in markets. The data has been there for years; the reasoning hasn’t.”

    Wn

    Working note

    AYVID Research · 2026

FAQ

Questions, answered.

Most of what we hear from allocators on the first call falls into one of these. The rest is in the operations console.

Ask another
Six asset classes across 300+ instruments on 30 exchanges: (1) Equities — long/short via shock-driven causal beneficiaries. (2) Index F&O — hedged macro exposure across 20 global indices. (3) FX — 34 pairs via macro-trend, carry, and mean-reversion. (4) Commodity futures — supply-shock, momentum, and seasonal edges. (5) Crypto — momentum and supply-chain expression on 22 assets. (6) ETFs & bonds — sector rotation, factor momentum, and rate curve trades. The operations console shows live positions and signal sources.
Causal reasoning over a live supply-chain graph. When a shock arrives, agents propagate it N hops to find unobvious beneficiaries and victims, then size the trade against historical co-movement. The edge is in the second-order chain, not the first headline.
No. Tier 1 (deterministic, < 100 ms) routes orders. Tier 2 (causal graph, ~1 s) re-weights conviction. Tier 3 (LLM swarm, ~30 s) can override — emit, modify, or cancel — but is never a blocker for the fast path.
AYVID is the buy-side production deployment of the anthropic/financial-services stack. We adopt the file-based Claude Skills convention (15 skill bundles ship in the repo: 10 adapted from the upstream equity-research and financial-analysis verticals, 5 AYVID-native), and we wire the buy-side slice of the MCP connector registry — MT Newswires, Aiera, FactSet, Moody's, S&P Global — directly into our continuous ingestion worker. Each connector is opt-in; the existing OpenBB / SEC EDGAR / news-scraper paths remain the default fallback. The full integration map lives in document/yc/anthropic_integration.md.
Position-level VaR, drawdown, concentration, and exposure limits are enforced pre-trade. An independent C++ kill-switch watchdog runs in its own process and can flatten the entire book in milliseconds if it detects anomalies.
Production engine trading 6 asset classes, IB Paper-traded, walk-forward backtested 2018→2025 across 20 named historical shocks. 300+ instruments on 30 exchanges, fully autonomous. We are running paper for the seed round and intend to convert to a regulated structure with first allocator capital.
We are selective by design. The fastest path is the contact form on /contact — please include AUM, mandate, and the typical allocation size you consider for early managers.

Engage with AYVID

Two ways in. Both start with a conversation.

Allocator track

Live operations review.

For LPs, family offices, and YC partners. Walk through the live shock catalog, signal book, and IB Paper trades. Sign an MNDA for equity-curve and risk-figure access.

Speak with the desk Operations console
Research track

Notes, methodology, and the engine roadmap.

Public research notes on causal inference, agent-swarm calibration, and walk-forward methodology. No live PnL.

Research notes Architecture