Proof of concept
An investment portfolio management system: trades, holdings, prices, dividends, performance and forecasts in one place, across several portfolios and several account types at once. A test bench too — an opportunity to put the Kore platform to the test on new ground: long time series, multiple currencies, and calculations that must be exact.
As soon as holdings sit across more than one account, "what is my overall position?" no longer has a simple answer.
Everything below works, and was checked against a fully synthetic demonstration dataset: fifteen invented securities, three portfolios, two years of daily history, three currencies. No real issuer, no real price, no real holdings.
The dashboard gathers the key indicators: number of lines, market value, gain, year-to-date return, and a score out of 100 computed by a fixed rule from return, volatility, maximum drawdown and number of lines. Below it, ten panels refreshed independently — concentration of the largest holdings, exposure by currency, a timeline of recent trades, alerts by state, top gainers and top losers, and a performance overview across three horizons.
A portfolio carries a strategy, an account carries a wrapper: the French PEA, PEA-PME, taxable brokerage, life insurance, PER, and plain cash accounts. The two attach freely, so a conservative sleeve and a satellite sleeve can share one account, or one strategy can span two brokers.
Thirteen security operation types and eleven cash movement types — buy, sell, subscription, redemption, deposit, withdrawal, fees, tax, interest, transfer, dividend. Holdings are never typed in: they are recomputed from the trades and the prices, cost basis and unrealised gain included.
A line quoted in dollars or pounds inside a euro portfolio is converted at the trade-date rate for its cost basis, and at the current rate for its valuation. The rate applied stays visible on every trade — a total whose calculation cannot be traced is not a reliable total.
Time-weighted return (TWR) and money-weighted return (MWR, via Newton-Raphson), volatility, Sharpe ratio, maximum drawdown, alpha and beta. All of it computed over a series of daily snapshots rebuilt by replaying the trade history, so it can be checked line by line. The performance page compares the portfolio with a reference index: return difference (alpha) and sensitivity to the index (beta).
Every distribution carries the quantity held at the date, the gross amount, the withholding, the net, and the credit on the cash account. The detailed list shows the security, the status (announced, pending, paid, cancelled) and the reference, and can be filtered, sorted and searched. The upcoming ex-date calendar and the forward yield projection build on the same history.
Daily closes and, where the source allows, intraday candles from one minute to one hour. Sources are configurable and prioritised; the origin of each price is kept, which is what tells a demonstration price from a market price — and makes the demonstration dataset removable in one command.
Reference instruments — exchange, sector, ISIN, currency — are shared and read-only. Each organisation adds its own layer on top: alias, private note, tags, chosen sector. No organisation edits the shared record for its own needs.
Moving averages, RSI, MACD, Bollinger bands, ATR, stochastic, ADX, CCI, Williams %R, OBV, MFI, VWAP, Chaikin: the indicators on price charts are computed over the history, not imported ready-made.
The screener filters securities on price, price change, volume, sector, currency, asset class and exchange, and on three technical indicators: RSI, MACD and position within the Bollinger bands. Screens can be saved and replayed.
A watchlist gathers the securities you follow without holding them. An alert fires when a threshold is crossed, up or down, or on a percentage move — with its priority, its channels, and the record of what triggered it.
The engine projects a median bracketed by its 10% and 90% quantiles, over four horizons, from one week to six months, with five interchangeable models — ensemble, GARCH, Theta, Holt-Winters, median. Replaying at past dates measures how well each one is calibrated: a forecast whose quality cannot be measured has no value. It is neither a price target nor investment advice.
A charts page plots portfolio value, unrealised gains, or the prices and volumes of followed securities over time, in whatever combination you choose. Three ready-made dashboards — portfolios, market, income — spare you from starting with a blank page. Only securities that have prices and portfolios that have a history are offered.
Built-in document management files account agreements, broker statements or key information documents into folders. A document can be attached to a portfolio, an account or a security, with a note saying why; the portfolio or security record shows its documents, and each document shows what it is attached to.
Nearly 290 GraphQL operations cover everything the interface does, and more. Three MCP tools and two assistant profiles let an agent read a portfolio or record a trade in it, with the organisation’s authentication and isolation intact.
Screens adapt to narrow widths and the application installs from the browser.
kFolio was first built to test the Kore platform on a domain where detail matters: long time series, several currencies, and figures a user checks on a calculator.
Users, roles and fine-grained permissions, isolation between organisations, translations, a single GraphQL API, an audit trail, centralised scheduling: kFolio inherits all of it instead of rewriting it. What remained was the business domain, which took most of the effort.
See the Kore platformA management screen tolerates a display approximation; a portfolio does not. Two totals diverging by 1.5% between two screens is a visible defect — and it is one that building the demonstration dataset brought to light and had fixed. An exploration that finds defects has already proved useful.
The application is hosted wherever you choose, price-provider credentials stay in your configuration, and the inventory of your holdings is stored in a database you own. Data providers are queried for prices, never told what you do with them.
kFolio is investment portfolio management software written by Koraisoft on its Kore platform. It tracks the trades, holdings, prices, dividends and performance of several portfolios, across French account types (PEA, PEA-PME, taxable brokerage, life insurance, PER), and shows probabilistic forecasts. It is a proof of concept: software explored internally, not a commercial offering.
No. kFolio is a Koraisoft proof of concept: it is not distributed, not supported, and no roadmap commits to it. There is no licence to buy, no trial version and no online service. If the subject interests you, getting in touch is the only way in today.
kFolio gives no investment advice and recommends no security. Its forecasts are probabilistic: a median bracketed by the 10% and 90% quantiles, over four horizons from one week to six months — a probability distribution, not a price target and not a recommendation. Replaying at past dates measures how well each model is calibrated, which tells you how it behaved, not what the market will do.
No. kFolio knows the account types (PEA, PEA-PME, taxable brokerage, life insurance, PER) and tracks withholding on dividends, but it produces no tax reporting: no gains computed under the tax method, no French flat tax (PFU), no 2074 form.
Trades are entered by hand, or imported from a CSV file or pasted text: an assistant proposes the column mapping, shows a preview and detects duplicates as the import runs.
The holdings, currency and performance formulas are held by the software rather than by hand, in a self-hosted, multi-user, multi-organisation web application with a GraphQL API, access for assistants and probabilistic forecasts. It is a proof of concept: it is not distributed.
Portfolio management, sovereignty over financial data, or more generally what can be built on the Kore platform: if any of these topics interests you, we would be glad to discuss it.
Get in touchThe screenshots on this page and the online demo run on fictional, deliberately small datasets: they illustrate only part of what the product does, and some features are limited or switched off in the demo environment.
