Serious investment research means reconciling many fragmented sources — quotes, filings, insider trades, analyst consensus, social and macro signals — into a single, current view. We built a terminal that does exactly that, then put an AI agent on top so the whole thing can be driven by conversation.
A resilient data layer
The system normalizes data across seven providers with automatic fallback, so a single outage never blinds the research loop. On top of the raw feeds it runs domain-specific detection — most notably Form 4 insider-cluster detection, surfacing the bullish pattern of open-market buying concentrated within a short window.
Conviction, not just data
Raw data isn't a decision. The terminal scores opportunities into a structured conviction view — combining quality, growth, and valuation signals with red-flag checks — so a candidate moves through a clear pipeline: generate → enrich → stress-test → score.
A sentinel that watches for change
A diff-based sentinel tracks what changed since the last scan — new insider clusters, analyst moves, fundamentals shifts — and emits those events to the agent, so the system surfaces what's new rather than re-reporting what you already know.
Agent-native by design
The capabilities are exposed as a tool layer (MCP) consumed by an agent built on the Claude Agent SDK. Research tasks — screening, building a dossier, checking insider activity, reviewing a position — become conversational operations backed by real, deterministic tools rather than guesses.
The interface
A full-screen terminal UI (built with Textual) presents equity curves, candlestick charts, and conviction scoring in a focused, keyboard-driven layout — the density of a professional research terminal without the per-seat price.
The project is a template for how we build AI products: a rigorous data and tool foundation first, with the language model as the interface to it — not a replacement for it.