Skip to main content
A market intelligence agent should separate market data, source evidence, and analysis. Prices and metrics describe what happened; filings and news provide primary context; social or prediction-market signals may help explain expectations but should not be treated as verified outcomes.

​
Choose the capability

​
1. Define scope

Specify the asset, company, market, event, and time horizon. Distinguish between:
  • Historical analysis.
  • Current-state monitoring.
  • Scenario analysis.
  • Forecasting.
A forecast should never be presented as a verified fact.

​
2. Collect structured data

Retrieve only the data required for the question, such as:
  • Prices and volume.
  • Financial metrics or statements.
  • Filings and filing sections.
  • Earnings or analyst information.
  • Relevant macro indicators.
Record the timestamp and units for every time-sensitive value.

​
3. Add primary-source context

Use filings, company releases, and official data when available. Use news to discover developments, then verify material claims against the most direct source possible.

​
4. Add optional market signals

Prediction markets and social discussion can indicate expectations, attention, or disagreement. Label them as signals rather than ground truth. Do not combine values from different sources without checking their definitions, currency, frequency, and observation time.

​
5. Synthesize with a model

Ask the model to separate:
  • Observed data.
  • Source-backed events.
  • Interpretation.
  • Scenarios and assumptions.
  • Risks and missing evidence.

​
6. Verify before delivery

  • Recheck prices and time-sensitive metrics.
  • Link filings or primary evidence.
  • State the cutoff time.
  • Mark forecasts and opinions clearly.
  • Avoid presenting the output as personalized financial advice.

​
Read versus write behavior

Market research is normally read-only. Do not place trades, alter portfolios, publish content, or trigger alerts unless the user explicitly requests that separate action and the relevant authorization path exists.

​
Minimal documentation set

​
When AIsa may not be necessary

A dedicated market-data provider may be simpler when the application only needs one dataset and already owns the analysis, model, and billing layers.