// system_overview

What is FAITH?

FAITH — Fully Autonomous Intelligent Trading Handler — is a modular AI-agent framework built for live on-chain trading on Solana. Each agent operates independently with its own wallet, strategy, and weighted memory system, allowing it to learn from every trade it executes and continuously refine its decision-making over time.

$how_it_works()

Agents

Each FAITH agent is an isolated execution unit. It holds a reference to a wallet, a trading strategy, and its own memory state. Agents run independently — one agent's decisions never pollute another's context.

Wallets

Wallets are Solana keypairs generated and stored via Privy's secure enclave. Private keys are never exposed to the FAITH backend. You link a wallet to an agent to authorize it to sign and submit transactions on-chain.

Execution

When an agent identifies a trade opportunity, it constructs a transaction, signs it using the linked wallet via Privy, and submits it to the Solana network. Execution latency targets sub-500ms from signal to on-chain confirmation.

Strategy Engine

Agents evaluate tokens using a configurable set of signals: price momentum, volume anomalies, liquidity depth, holder distribution, and real-time social sentiment pulled from X and Telegram. Each signal is scored and weighted to produce a composite trade confidence score.

$social_intelligence()

Human traders lose to emotion. FAITH doesn't. Instead of reacting emotionally to noise, FAITH ingests real-time data from social platforms and converts crowd sentiment into a quantified signal — one input among many, weighted by historical accuracy, never the sole reason to trade.

Live Social Feeds

FAITH monitors real-time activity across X (Twitter) and Telegram — the two platforms where Solana token narratives move fastest. Posts, mentions, and volume spikes are ingested continuously and scored for signal quality.

Sentiment Scoring

Raw social data is processed into a normalised sentiment score per token. This is not keyword matching — it is a weighted model that distinguishes genuine accumulation narrative from coordinated shill campaigns.

Velocity Detection

A sudden spike in mention volume before a price move is one of the most repeatable patterns in meme token markets. FAITH tracks mention velocity as a leading indicator, not a lagging confirmation.

Emotion Removed

Human traders buy into hype and panic-sell into dumps — classic emotional decision-making. FAITH treats social sentiment as data: a signal with a known accuracy rate, not a feeling to act on impulsively.

Cross-Signal Fusion

Social sentiment never triggers a trade alone. It is fused with on-chain signals — liquidity, volume, holder distribution — and the agent's memory model. All inputs must align before confidence crosses the execution threshold.

Faster Than Human

By the time a human sees a trending token, processes the sentiment, and decides to buy, FAITH has already evaluated the signal, scored it against its memory, and either acted or dismissed it — in milliseconds.

The edge is not more information — it is better processing of the same information.

Every trader on Solana sees the same tweets, the same Telegram pumps, the same X narratives. The difference is that FAITH processes this data without fear, greed, or FOMO — applying a consistent, memory-weighted scoring model that has been calibrated against real trade outcomes, not gut feeling.

$agent_memory_system()

Every agent maintains a private weighted memory store. This is not a fixed ruleset — it is a living model of the agent's trading history that evolves with each outcome.

Trade Reflection

After each closed position, the agent reflects on the trade: what signals were active, what the outcome was, and how much confidence it should assign to that pattern going forward.

Strategy Weighting

Strategies are not binary on/off switches. Each strategy carries a confidence weight from 0.0 to 1.0. Repeated wins in a strategy increase its weight; repeated losses decay it — following an exponential moving average.

Adaptive Confidence

Over time an agent builds conviction in the patterns that consistently work for the tokens and market conditions it operates in. Low-confidence strategies are deprioritised automatically, not discarded.

Pattern Scoring

Each observed market pattern — breakout, accumulation, volume spike, liquidity pull — is scored against historical outcomes stored in the agent's memory. The score informs position sizing.

Memory Isolation

Memories are scoped strictly to the agent. Two agents running the same base strategy will diverge over time as each builds its own independent confidence model from its own trade history.

Confidence Threshold

Agents only enter a trade when the composite confidence score — memory weight × signal strength — crosses a configurable threshold. This prevents low-conviction noise trades from eroding capital.

// confidence_weight_formula

W(t+1)=α × outcome(t) + (1 − α) × W(t)

W(t) — current strategy confidence weight

outcome(t) — normalised result of the last trade (0.0 loss → 1.0 win)

α — learning rate (default 0.15, configurable per agent)

A win increments weight toward 1.0 at rate α. A loss pulls it toward 0.0. The agent never fully commits to or abandons a strategy — it converges on what empirically works in live conditions.

$security_model()

Privy Secure Enclave

All wallet private keys are generated and held exclusively inside Privy's secure enclave. The FAITH backend never receives, stores, or logs raw private keys at any point — signing requests are delegated to Privy directly.

Zero Key Exposure

FAITH operators — including the development team — have zero access to your wallet private keys. The enclave architecture is designed so that key material cannot be extracted even with full server access.

$free_tier()

FAITH is free to use with no balance required.

Agents on the free tier are fully functional but operate under a shared execution queue. Without a balance, your agent's priority is determined by a first-come-first-served scheduling model — meaning scan intervals may be delayed, execution latency will be higher, and market scan frequency may be significantly slower than listed target intervals.

—Scan intervals may be inaccurate or delayed under high system load.

—Execution speed is not guaranteed on the free tier — opportunities may be missed.

—Priority is given to agents with a funded balance. Free agents fill remaining capacity.

—All other features — memory, strategy weighting, trade history, leaderboard — are fully available at no cost.

// FAITH is experimental software. Trade only what you can afford to lose.