Documentation Index
Fetch the complete documentation index at: https://mintlify.com/manusapis/Agix/llms.txt
Use this file to discover all available pages before exploring further.
Agix ships three analytics services that operate directly on your spreadsheet columns: forecastService projects future values from historical date-indexed data, anomalyService flags rows that fall outside expected statistical bounds, and sentimentService scores text entries and tracks satisfaction trends over time. All three can be combined with the scheduler to run automatically on a recurring basis.
Forecasting
forecastService lives at src/services/analytics/forecast.service.ts and exposes three methods for time-series work.
predict(request)
Generates a forward projection for a numeric column indexed by dates.
async predict(request: ForecastRequest): Promise<ForecastResult>
// ForecastRequest
interface ForecastRequest {
dateColumn: string; // Column letter or header name
valueColumn: string;
horizon: number; // Number of future periods to predict
seasonality?: "daily" | "weekly" | "monthly" | "yearly" | "auto";
}
// ForecastResult
interface ForecastResult {
forecast: ForecastPoint[]; // One entry per future period
confidenceIntervals: ConfidenceBand[]; // e.g. 95% band
modelUsed: string; // e.g. "seasonal-decomposition"
influentialPoints: InfluentialPoint[];
}
// ForecastPoint
interface ForecastPoint {
date: string; // ISO date string
predicted: number;
lowerBound: number;
upperBound: number;
}
const result = await forecastService.predict({
dateColumn: "A",
valueColumn: "B",
horizon: 30, // predict the next 30 days
seasonality: "weekly",
});
result.forecast.forEach((point) => {
console.log(`${point.date}: ${point.predicted} [${point.lowerBound}–${point.upperBound}]`);
});
decompose(dates, values)
Breaks a time series into its structural components for inspection or charting.
decompose(dates: string[], values: number[]): {
trend: number[];
seasonal: number[];
residual: number[];
}
findInfluentialPoints(dates, values)
Returns the historical data points that most heavily shaped the forecast model.
findInfluentialPoints(dates: string[], values: number[]): InfluentialPoint[]
// InfluentialPoint
interface InfluentialPoint {
date: string;
value: number;
influence: number; // relative weight in the model, 0–1
}
Anomaly Detection
anomalyService lives at src/services/analytics/anomaly.service.ts. It accepts a numeric column and a detection method, then returns every row that falls outside expected bounds.
detect(columnValues, columnName, method)
async detect(
columnValues: number[],
columnName: string,
method: "zscore" | "iqr" | "isolation-forest"
): Promise<AnomalyResult>
Detection algorithm to use. Defaults to "zscore" if omitted.
Z-Score
IQR
Isolation Forest
Computes the mean and standard deviation of the column, then flags any row where the absolute z-score exceeds 2.5.z = |( value − mean ) / stddev|
flagged when z > 2.5
Computes Q1 and Q3, then flags values outside the fences:lower fence = Q1 − 1.5 × IQR
upper fence = Q3 + 1.5 × IQR
Routes to the z-score implementation in the current release. Full isolation-forest support is on the roadmap.
Result type
interface AnomalyResult {
column: string;
anomalies: AnomalyPoint[];
threshold: number; // 2.5 for zscore, 1.5 for iqr
method: "zscore" | "iqr" | "isolation-forest";
}
interface AnomalyPoint {
rowIndex: number;
value: number;
expectedValue: number; // mean (zscore) or midpoint of Q1+Q3 (iqr)
deviation: number; // z-score or absolute distance from expectedValue
}
const revenue = [12000, 13400, 11800, 98000, 12700, 11200]; // 98000 is suspect
const result = await anomalyService.detect(revenue, "Revenue", "zscore");
result.anomalies.forEach((pt) => {
console.log(`Row ${pt.rowIndex}: ${pt.value} (deviation: ${pt.deviation.toFixed(2)})`);
});
// → Row 3: 98000 (deviation: 3.47)
Sentiment Analysis
sentimentService lives at src/services/analytics/sentiment.service.ts. It scores every text entry in a column and optionally tracks how sentiment shifts across multiple analysis runs.
analyze(columnValues, columnName)
Scores each row as "positive", "negative", or "neutral", and returns an overall column score.
async analyze(
columnValues: string[],
columnName: string
): Promise<SentimentResult>
interface SentimentResult {
column: string;
entries: SentimentEntry[];
overallScore: number; // average score across all rows, −1 to 1
trend: "improving" | "declining" | "stable";
}
interface SentimentEntry {
rowIndex: number;
text: string;
score: number; // −1 (most negative) to 1 (most positive)
label: "positive" | "negative" | "neutral";
}
- Customer feedback or NPS comment columns
- Product reviews imported from a web source
- Support ticket subject lines or descriptions
- Employee survey free-text responses
trackTrend(columnName, history)
Compares the last three SentimentResult snapshots in history and classifies the direction of change.
async trackTrend(
columnName: string,
history: SentimentResult[]
): Promise<"improving" | "declining" | "stable">
history array. If the most recent overallScore is more than 0.1 higher than the oldest of the three, the trend is "improving". If it drops more than 0.1, it is "declining". Otherwise it is "stable".
Example:
const feedback = [
"Great product, really happy with it.",
"Support was slow to respond.",
"Exactly what I needed.",
];
const result = await sentimentService.analyze(feedback, "CustomerFeedback");
console.log(`Overall: ${result.overallScore.toFixed(2)} — ${result.trend}`);
result.entries.forEach((e) =>
console.log(` Row ${e.rowIndex}: [${e.label}] ${e.text}`)
);
All three analytics services integrate with the scheduler. You can register a daily refresh-analysis task to re-run forecasts, anomaly checks, or sentiment scans automatically as new data lands in your sheet. See the Scheduler page for a step-by-step example. 