DCF
Discounted Cash Flow
Present value of expected free cash flows after discounting them by the cost of capital.
Best when: Stable, predictable cash flow.
Equity, analyzed
The name is equity plus analyze, which is pretty much the whole idea. Give Equilyze a company and it runs the valuation engine, checks market signals, reads recent news, and asks a local language model to turn the pieces into an investment report.
The result is one structured report with a company overview, financials, valuation detail, a recommendation, follow-up sections, and a final verdict. Everything runs against an Ollama model on the local machine, so no cloud API keys are needed.
From a ticker to a finished report
You type a company name or symbol. Equilyze resolves the ticker, runs the valuation engine, then hands the numbers to a chain of agents that add context, research the news, write a recommendation, ask their own follow-up questions, and review the report before saving it.
Flowchart
Six valuation models, three scenarios each
Equilyze runs the Equity Valuation Engine across six valuation models. Each model is estimated under Bear, Base, and Bull scenarios. The engine then computes a composite intrinsic value and a model agreement score that shows how far the models disagree.
PE
Price / Earnings
Fair price implied by applying a target P/E multiple to current earnings.
Best when: Mature, profitable companies.
ROE
Return on Equity
Intrinsic value based on the company's return on shareholder equity.
Best when: Capital-efficient businesses.
EV/EBITDA
Enterprise Value / EBITDA
Operating value before capital structure gets in the way.
Best when: Cross-sector and M&A comparisons.
PS
Price / Sales
Revenue-based valuation, useful when margins are low or volatile.
Best when: Early-stage, high-growth companies.
NAV
Net Asset Value
Liquidation value after subtracting total liabilities from total assets.
Best when: Asset-heavy industries.
Six agents, running one at a time
The valuation numbers feed a sequence of agents. Some gather context and news. Others turn that material into sections, ask follow-up questions, and check the final report.
Agent 1
Contextualizer
Fetches analyst ratings, price momentum, and earnings, then writes the market context in plain bullets.
Agent 2
News & Sentiment
Pulls yfinance headlines plus two web searches and summarizes the news around the company.
Agent 3
Synthesizer
Turns the valuation, context, and news into a recommendation with strengths, risks, and catalysts.
Agent 4
Curious Investor
Reads the draft and asks the follow-up questions an investor would probably still have.
Agent 5
Section Writer (×3)
Answers each question as its own report section, grounded in the valuation, analysis, and news data.
Agent 6
Final Reviewer
Reads every saved file from disk and appends a verdict with the closing sections that fit the report.
How a report is assembled
- Ticker
- Valuation
- Contextualize
- News
- Recommend
- Questions
- Sections
- Review
Every run leaves four artifacts behind
The report is only the visible output. Under reports/TICK/, Equilyze also saves the structured data each agent worked from. Those artifacts make the analysis auditable, and they let another tool pick the work up and carry it further.
report.md
The human-readable report
The full Markdown report with the overview, financials, valuation section, recommendation, and verdict.
valuation_data.json
Raw engine output
Every metric, per-model intrinsic value, scenario, composite, and agreement score from the valuation engine, kept exactly as computed.
analysis.json
Structured recommendation
The StockAnalysis object with recommendation, confidence, strengths, risks, catalysts, and narrative fields.
news_data.json
Researched news
The raw headlines and web-search results gathered for the company, so the sentiment in the report can be traced back to its sources.
Case study, from Oracle analysis to article
Is Oracle Stock Undervalued? began here. Equilyze evaluated Oracle and produced its report and artifacts. Those files then became the input for Pressroom, a nine-agent pipeline that wrote the finished article.
Flowchart
Equilyze ran first and wrote report.md, valuation_data.json, analysis.json, and news_data.json for Oracle.
Those four artifacts were handed to Pressroom as source material. Its agents outlined the piece, drafted it, checked it against the JSON, rendered the charts, and published the final .mdx. The valuation numbers in the article are the ones this engine computed.
Configured for local, memory-aware inference
Environment
Point it at a model and the engine
A short .env file sets OLLAMA_MODEL, OLLAMA_BASE_URL, a timeout, and EQUITY_ENGINE_PATH, which points to the valuation engine's src/ directory.
Engine invocation
In-process first, subprocess as fallback
Valuation runs the engine in-process via sys.path injection for speed. If that fails, it falls back to running the engine's CLI as a subprocess and captures its JSON.
Inference
Sequential on purpose
Every model call runs one at a time. Parallel inference with a 20 GB+ model on a 36 GB machine causes memory pressure, so Equilyze spends more time to avoid crashes.
The outcome
Equilyze pairs a six-model valuation engine with a chain of agents to turn a single ticker into a report that shows its work.
Quantitative models set the anchor. Agents add context, challenge the draft, and sign off using the JSON snapshots saved next to every report. The report closes with a disclaimer, because Equilyze studies companies. It does not give advice.