How it works

Analyst Agent is an AI-powered SEC filing analysis platform. Enter any public company ticker and ask natural language questions — the agent fetches SEC filings, runs LLM analysis, and streams results back in real time.

The Story

Analyst Agent is a comprehensive AI-powered financial analysis platform that turns SEC filings, market data, and web research into natural-language answers.

The architecture evolved from a simple ReAct agent to an intelligent planning system that routes queries based on complexity — using dynamic tool selection for simple questions and structured multi-step execution for complex financial analysis. Independent analysis steps run in parallel to minimize response time.

Most recently, the platform expanded beyond chat-only interaction with a Company Dashboard that proactively surfaces SEC filing analysis, technicals, and company metadata the moment a user selects a ticker — no prompt required.

Graph Architecture

START
  └─▶ [ROUTER] ── classifies query complexity
        ├─▶ simple  ──▶ [REACT AGENT] ─────────────▶ END
        ├─▶ unclear ────────────────────────────────▶ END
        └─▶ complex ──▶ [PLANNER]
                          └─▶ dispatch (Send fan-out)
                                ├─▶ [WORKER] ┐
                                ├─▶ [WORKER] ├─▶ [RECONCILER] ──▶ [SYNTHESIZER] ──▶ END
                                └─▶ [WORKER] ┘   (independent steps, in parallel)

The agent classifies query complexity to pick an execution path. Simple queries are handled by a ReAct agent with dynamic tool selection. Complex queries are decomposed by a planner into independent steps that fan out to parallel workers (LangGraph Send); a reconciler merges their results and the synthesizer assembles the final response.

3-Tier Caching

1edgartools Company object· Per process lifetime

2LLM analysis (Pydantic models)· Per accession number (SQLite)

3Tavily web search results· Per query hash (in-memory)

Capabilities

SEC Filing Analysis

13 tools

Powered by edgartools. Raw and AI-analyzed 10-K/10-Q sections — risk factors, MD&A, balance sheets, full 10-K text, business overview, cybersecurity disclosure, legal proceedings, and a concurrent all-summaries tool. 8-K support with earnings release analysis (beats/misses, guidance) and material event analysis. Supports 20-F for foreign filers.

Market Data & Technicals

7 tools

Via Yahoo Finance. Price history, stock info (P/E, market cap, beta), financial metrics (revenue/income growth, debt ratios), core technicals (RSI, MACD, Bollinger Bands, 5/10/20/50/200-day MAs), advanced technicals (ADX, ATR, Stochastic, Volume Profile, Fibonacci), chart pattern detection, and multi-timeframe analysis with conflict detection.

Market Overview

2 tools

Current levels and daily changes for S&P 500, Nasdaq, Dow Jones, and VIX. Macro indicators including 10-Year and 5-Year Treasury yields, VIX, and US Dollar Index with 1-month changes.

Web Research

5 tools

Optional tools via Tavily. General web search, deep research (async polling), company news, competitor analysis, and industry trend forecasting. Requires a Tavily API key.

Briefing History

2 tools

Retrieve recent briefing analyses for a ticker (outlook trends, technical signals, news summaries) or the latest full morning briefing (market regime, positioning, alerts).

Company Dashboard

Instant Overview

Selecting a ticker loads a metrics grid, technical snapshot, chart patterns, and market regime badge instantly — no LLM call, no prompt. Backed by a 5-minute server-side cache.

LLM-Analyzed Filings

The Filings tab streams AI analysis of the latest 10-K (or 20-F), 10-Q, and 8-K earnings release via SSE. All analyses run concurrently and render incrementally as they complete. Cached in SQLite by accession number — auto-invalidates when a new filing appears on EDGAR.

Integrated Chart + Chat

The third tab embeds the full interactive chart (TradingView Lightweight Charts with RSI, MACD, Bollinger Bands) alongside the conversational agent — 29 tools available when Tavily is configured (24 without).

API Key Architecture

Keys are never stored on the server. Bring-your-own-key values — for any of Google Gemini, OpenAI, or Anthropic models — are saved in your browser's localStorageand injected per-request as HTTP headers or WebSocket auth messages. Anonymous visitors can browse and analyze with their own key; signing in (Clerk) unlocks saved sessions and watchlists. The server's own environment keys are lent only to the local/self-host operator or verified signed-in users — never to anonymous traffic — so drive-by requests can't spend them.