AI Transparency

Who we are

Djembely is owned and run by Melian Dialogue Limited (the "Company"). It is a private limited company set up in the Republic of Kenya.

Registered Office: P.O. Box B1003, Urban Oasis Apartments, Kingara Road, Lavington, Nairobi.

Company Registration No.: PVT-27ULY98K.

KRA PIN: P052416974R.

Code first, AI second

Djembely works out every basic fact using ordinary code. It uses your own telemetry (information about how you use the app while you are signed in) and market data. The market data comes from EODHD, Mansa Markets, Coinpaprika and Alpaca (paper-trading data).

We never ask the AI model to invent an analysis. We only ask it to put facts that already exist as data into words. It does this in the language and level of detail you choose.

This keeps costs down. It also lowers the risk of the AI "making things up".

Worked example: a pre-trade read

Say you open the Co-pilot's pre-trade analysis. Ordinary code works out all of these things. This is called DETERMINISTIC, which means the same inputs always give the same answer:

• ·Your stop distance compared with the ATR band [0.5×ATR, 4.0×ATR]. (A stop is a point where you plan to sell to limit a loss. ATR, or average true range, shows how far a price usually moves.)
• ·Your position size compared with your target risk.
• ·Which technical-analysis idea the current price is actually touching, for example an earlier resistance level (a price where it often stops rising).

What the AI writes

The AI only writes the sentence around those numbers. For example, it can turn "stop = 3.1×ATR, in band" into plain words at Beginner, Intermediate or Technical level.

On the Cadet tier there is no AI call at all. You see the same facts with fixed, ready-made wording.

Post-trade analysis works the same way. Code decides which of the four process/outcome quadrants a closed trade falls into, using that trade's own BeCo values. The explanation around it is written by AI for paid tiers. Cadet gets fixed wording.

Model and safeguards

The Co-pilot uses OpenAI GPT 6 Luna. We use it through the Company's own OpenAI account. If that account is not available, the Emergent universal key takes over automatically as a backup.

You can also type or speak a trade (prompt-to-trade). Voice uses OpenAI speech-to-text (turning your speech into words) and text-to-speech (reading words aloud).

There are daily limits on how much the AI can do for you. Cadet gets 0 live AI calls, because it uses fixed wording only. Trader gets 50 AI-written analyses per day. Professional gets 150 per day. There is also a limit of 4 in-trade calls for each position. These are the only limits in force across the whole product right now.

The Co-pilot never blocks you from submitting a practice trade. It never pushes you to trade or spend more. It does not guess your feelings from your voice or body signals.

There is one exception, and it comes from the trading rules, not from the AI. A short sale (selling something you do not own) must be funded first with collateral (money set aside as a guarantee). The collateral is the larger of 25% of the position or GBP 1,000 at today's rate. Short sales without this funding are declined.

AI answers can be wrong. They are for learning only. They are never financial advice, and they never promise returns.

Autotrader and Auto-Discovery

Autotrader is practice-only automated trading in up to 5 slots. Auto-Discovery runs scans on weekdays on a schedule. It can suggest or add instruments, but only within your settings.

Each decision combines several signals into one confidence score. The signals are:

• ·Price and volume (how much is being traded).
• ·EODHD financial news and sentiment (whether the news sounds positive or negative).
• ·Company events: dividends, earnings dates, price-to-earnings figures and profit figures.
• ·The EODHD economic calendar, such as interest-rate decisions and inflation.
• ·For African shares, company announcements from Mansa Markets.

The rationale log

Every Autotrader trade and every recommendation has a rationale log. It is written in plain English. It shows each signal, how much weight it was given and which data source it came from.

Backtests use past prices only. They do not predict the future.

Data, translations and limits

Market data can be delayed or wrong. For African exchanges, Mansa Markets can give multi-year price history where their plan allows. Short gaps of up to 3 trading days are filled in and flagged.

Djembely is available in seven languages: English, Spanish, French, Swahili, Sheng, Luganda and Kinyarwanda. The AI drafts the translations. If a translation and the English version say different things, the English version wins.

Djembely is for practice trading only, with virtual money (GBP 10,000 or a regional equivalent). It has no real orders and no leverage (borrowed money to make bets bigger). Learners must be 18 or over.