A private agent that runs on your own hardware. It pulls live market data, reads official filings, runs deep research, models valuations, gauges sentiment, and checks its own facts. Your holdings never leave the house.
Curunir has memory, a personality, and its own tools, so it works more like a colleague than a one-shot answer box. It keeps track of your holdings and your cost basis, remembers the theses you're working between conversations, and picks the right tool on its own.
It works alongside you on the whole book: trading and speculation, long-term investing, retirement, and tax. It tracks your positions, builds and stress-tests theses with you, and backs every claim with a source you can open.
And it does all of it in your house. The language models live in the cloud; your portfolio and your notes stay on the device, in local memory.
A thesis is only as good as what it stands on. Every answer here is grounded in a real source: filings, market data, macro series, prediction-market odds, social chatter. Each one is pulled live and cited inline, then handed to a separate agent that fact-checks it before you see it. No vibes, no made-up numbers.
And the roster keeps growing. We add skills, tools and integrations all the time, and each new one is simply another source the agent can reach for and cite. You ask the same way you always did.
Your balance sheet comes from a real engine, not from prose the model writes. Holdings are grouped by class, with a net-worth rollup, a specific-lot trade ledger that tracks realized and unrealized P&L, and point-in-time snapshots you can pull up and compare. Every figure is computed, so the agent can't quietly hand-sum a total or make one up.
Export a CSV from your brokerage and the engine folds those positions straight into your book, cost basis and lots included, in a single step. Live account sync is on the roadmap. For now the CSV import keeps your holdings current, and your login stays with you.
Build a thesis as a fact-checked memo, stress-test the bull and bear cases, then health-check it again as the facts move. Your notes live in local memory, so re-running the memo or the fact-checker months later picks up where you left off, corrections and all.
Deep, multi-source research: it breaks the question down, pulls from authoritative sources like SEC EDGAR filings, yfinance market data and FRED macro series, then hands the draft to an independent fact-checker. Every claim traces back to a source you can open yourself.
Read your allocation off the net-worth rollup, then work lot by lot with tax in mind: tax-loss harvesting, the wash-sale rule, short- vs. long-term holding periods. The guidance comes from this year's IRS rules, which the agent looks up fresh each time instead of recalling from memory.
SPCX targets a June 12, 2026 Nasdaq debut at a $1.75–2 trillion valuation — the largest IPO in history. I built both cases straight from the S-1.
Twelve sections, fact-checked and corrected. Full PDF attached.
Retatrutide (LY3437943) is Lilly's triple agonist. It posted 28.7% average weight loss at 68 weeks in Phase 3 — the highest ever recorded, ahead of tirzepatide and semaglutide.
Stock impact: LLY trades around $963; approval could drive $1,200–1,300 (consensus ~$1,220), a setback −15–25%. Memo attached.
I checked it before agreeing with it — and the data didn't back the inflation-hedge claim. Bitcoin is trading risk-on right now, with roughly a 0.74 correlation to the S&P. I said so plainly rather than confirm what you already believed.
You pivoted to gold, so I built that thesis in the same format — central-bank buying, debasement, and the real-rate backdrop — fact-checked, with three corrections applied before it reached you.
Pulled the fund facts, benchmarked the returns, and stress-tested each pillar of your thesis. The pillars hold up — no fundamental rethink needed.
Where the evidence was uncomfortable, I showed it rather than smoothing it over.
The finance skills above are one way to configure Curunir, a small agentic harness we kept deliberately bare. The same engine that writes an investment memo can run research, handle scheduling, or do whatever else you wire into it.
It runs on a small device in your home or office. The language models live in the cloud — but your portfolio, your cost basis, and your thesis notes don't. They stay on the device, in local memory.
You can reach it from anywhere — web, CLI, or email — through a thin portal that passes the conversation along and keeps the data where it is.