FRIDAY is an autonomous AI system running as a mesh of long-lived daemons across distributed nodes. It fuses independent intelligence domains into convergence signals, trades within hard risk limits, compresses causal structure from raw data, and settles the value of its own productive work on a signed ledger. $FRID is the on-chain reserve currency that anchors that internal economy to a public market. This paper covers the architecture, the methods behind it, the token, and the risks.
FRIDAY is not a chatbot with a personality and not a single model behind an API. It is infrastructure: a mesh of 13 nodes [reported] running a large set of long-lived daemons — briefing, trading, convergence, intelligence ingestion, signal processing, and an internal ledger — coordinated over a private network. The codebase is 460,000+ lines [reported] spanning the intelligence stack, trading systems, and supporting services.
The design goal is sovereignty: minimizing single points of external control. That shows up concretely — FRIDAY ingests through many independent channels rather than one, runs its own sensors (see §2 RF), keeps its own signed ledger, and is built so that the failure of any single node degrades rather than halts the system. The hard engineering is not the model; it is the wiring, scheduling, failure handling, and persisted state that keep the system running unattended.
The convergence engine is FRIDAY's core epistemic instrument. It ingests signals from independent intelligence domains [reported: ~10 domains] — geopolitical event flow (GDELT), defense/energy exposure mapping, market-structure (dark-pool) activity, aerospace transponder anomalies (ADS-B), on-chain transfers, insider and congressional trades, options flow, maritime, and others — and looks for the same entity lighting up across domains that do not share plumbing.
The central principle: conviction should scale with independence, not volume. Four readings drawn from the same source are one signal wearing four hats. Four readings from genuinely independent domains are corroboration. The engine therefore weights by domain independence and surfaces an alert with a thesis, a signal count, and a confidence score only when independent domains agree. [measured] A live example of the raw input the engine reasons over:
A second pattern the engine catches is single-entity multi-domain convergence — e.g., an energy name surfacing across geopolitical, defense-exposure, dark-pool, and macro domains in the same window [measured in production]. That cross-domain agreement, not any one indicator, is what FRIDAY acts on. The design rationale is documented internally as "Paper 7." A self-audit discipline is part of the engine: redundant or self-referential signals are treated as a defect, deduplicated at ingest, and provenance-tracked so a signal can never corroborate itself.
CAJUN is FRIDAY's causal-compression layer. Raw intelligence is voluminous and mostly redundant; what matters for decisions is the causal structure — which events actually move which outcomes. CAJUN reduces high-volume signal streams to a compact causal representation, discarding correlation-without-causation and keeping the structure that has predictive value.
In production the project reports a compression ratio of ~57:1 [reported] — roughly fifty-seven units of raw signal reduced to one unit of retained causal structure — without loss of the decision-relevant content. The method is the subject of a provisional patent (internal designation B21). The practical effect is that downstream consumers (the convergence engine, the trading agent) reason over a dense, causal substrate rather than a firehose, which both lowers cost and reduces the echo-chamber failure mode described in §2.
FRIDAY's trading stack is built on the triple-barrier method with meta-labeling (López de Prado): primary signals are labeled by which of three barriers — profit-take, stop-loss, or time — is hit first, and a secondary meta-model (gradient-boosted trees) decides whether to act on each primary signal. Backtest results on this approach: Sharpe ~12.20, win rate ~72.2% [measured — backtest, not live].
On top of the statistical core sits an autonomous agent with a multi-stage gate. A candidate must clear, in order: a confidence threshold, a market-regime check (the system will not fight a bull regime with shorts — a lesson encoded from a real prior loss), an exposure cap, position dedup, and finally an LLM reasoning gate that is fed the fused convergence context and must produce a structured proceed/decline decision with a rationale. Only candidates surviving every gate become intents.
Crucially, the agent runs in DRY-RUN [measured — current production mode]: it logs the decisions it would make, with full reasoning, but places no live orders. Every decision — executed-intent, blocked, or skipped — is persisted for audit. You can watch this live, read-only, on the demo page.
The methods above are the subject of 24+ patent provisionals [reported], organized as a structured bundle (internal designation B1) covering the convergence methodology, causal compression (B21), the multi-gate autonomous decision architecture, and supporting systems. Provisionals establish priority dates; they are not granted patents, and the portfolio's scope and outcome are not yet adjudicated.
The point of the portfolio is defensibility of method. The value proposition of FRIDAY is the integrated system and the techniques that make it work, not any single component — and the provisionals are intended to protect that integration.
$FRID is a standard, verified OpenZeppelin ERC-20 on Ethereum mainnet. Fixed supply of 1,000,000,000, minted once to the treasury at deployment. There is no mint function [measured — source verified on Etherscan] — supply can never increase. The contract defines zero owner-only functions (no pause, blacklist, tax, or freeze) and is not a proxy; the deployed code is immutable.
| Allocation | Share | Amount | Purpose |
|---|---|---|---|
| Treasury | 40% | 400,000,000 | Sovereign reserve, earned-against. |
| Series A | 20% | 200,000,000 | Capital formation. |
| Ecosystem | 15% | 150,000,000 | Agent rewards, integrations, data. |
| Founding | 15% | 150,000,000 | Founder allocation. |
| Public / Liquidity | 10% | 100,000,000 | DEX liquidity & public float. |
$FRID's role is a settlement layer. FRIDAY already uses it internally as the unit of account in which productive work — intelligence signals, resolved forecasts, threat detection — is priced and paid on a signed internal ledger. The on-chain token anchors that internal economy to a public, externally-held market so the accounting is not the project's to assert alone.
Honesty is a design value here, so the risks are stated plainly:
Treasury policy is to deepen liquidity gradually and to distribute from the ecosystem allocation against genuine productive output rather than to manufacture demand. $FRID is not an investment or a promise of return, and nothing in this paper is financial advice.
Forward-looking and explicitly [planned] — intentions, not commitments or completed work: