How it works
One agentic engine powers both the dashboard and the chat. Here is what happens between your question and a cited answer.
Manager → Delegator → Workers
The Manager agent turns your request into a plan of tasks. The Delegator schedules them as a dependency graph. Worker agents execute each task in an LLM + tool loop, calling specialized tools and returning grounded results.
Eight registered tools cover the analyst's workflow: extracting financial data, running analyses, computing ratio sets, charting, pulling price history, estimating valuations (DCF and dividend models), searching the document corpus, and generating multi-document answer grids.
Three data planes, one normalized corpus
Adapters normalize every source into a common schema with provenance attached, so the agents reason over one consistent corpus.
External
SEC EDGAR full-text filings and XBRL company facts, market data (Alpha Vantage, Polygon, Yahoo Finance), news (NewsAPI, GDELT), earnings-call transcripts, and alternative data such as web traffic and app metrics.
Internal
Adapters for your SQL warehouse and REST services bring internal financials and operating metrics into the same analysis, normalized to the same schema.
Uploads
CSV, Excel, PDF, and text files are parsed, chunked, and embedded into the corpus, so an uploaded statement is as searchable and citable as a public filing.
Traceable by construction
Transparency isn't a report generated afterwards — it's how the system runs.
Visible reasoning
The plan, the task DAG, and every tool invocation stream to the UI live. You can audit how a conclusion was reached, not just read it.
Stored provenance
Citations aren't decorative — every figure and statement links to a stored source line item or document chunk.
Accounted compute
Each run records its LLM token usage and external API calls, and budgets are enforced before costs run away.