NexisQ doesn't try to predict the market. It audits the quality of execution of a trade idea you already have — timing, risk, structure, discipline.
You describe a trade in plain language ("BTC long 10x on 15m, market entry, no stop, risk 5%, deposit 1000"). The bot extracts a structured idea, asking follow-up questions with inline buttons for anything missing, pulls a live market snapshot from BingX, computes technical indicators across two timeframes, and runs it all through an LLM that returns a sober, plain-text breakdown: is the entry sane, where are the weak spots, is the risk adequate, is it too late to enter.
Access is gated behind a crypto subscription paid via Crypto Pay (@CryptoBot). The whole thing runs on long polling — no domain, no webhook, no TLS — so it deploys as a single Docker container.
Sped up · full flow: idea → follow-up questions → market snapshot → AI verdict
- 🧠 Execution audit, not signals — evaluates your idea instead of inventing one.
- 💬 Natural-language input — an LLM parses a free-form message into a structured trade; missing fields are collected with inline keyboards.
- 📊 Real market context — live BingX klines, funding rate, open interest and 24h ticker, with RSI / EMA / SMA / ATR / swing levels / volume trend computed on both the working and a higher timeframe.
- 🔌 Pluggable LLM — Anthropic or OpenAI out of the box; any OpenAI-compatible gateway (e.g. OpenRouter) via a custom base URL.
- 💳 Crypto subscriptions — Crypto Pay invoices with idempotent activation (background reconcile + a manual "I paid" button, no webhook needed).
- ⚙️ Live-editable from the bot — texts, the AI instruction, price, currency, duration and model names are changed via
/textswith no restart. - 🛟 Operational care — admin alerts on AI/exchange failures (anti-spam), a daily stats digest, throttling, a global error handler and graceful startup without an AI key.
- 🧪 Tested core — pure indicator functions covered by unit tests; SQLite with WAL and file-based migrations.
flowchart TD
A[User sends a trade idea<br/>as free-form text] --> B[LLM parses it into<br/>a structured TradeIdea]
B --> C{All fields<br/>filled?}
C -->|No| D[Ask the missing field<br/>via inline keyboard]
D --> C
C -->|Yes| E[Fetch BingX snapshot:<br/>klines · funding · OI · ticker]
E --> F[Compute indicators on<br/>main + higher timeframe]
F --> G[LLM analyses idea + snapshot]
G --> H[Plain-text verdict + disclaimer<br/>sent to the user]
H --> I[(Saved to<br/>analysis_history)]
Two LLM calls per analysis: a cheap model parses the idea from natural language, a stronger model analyses it. Both model names are configurable at runtime. For the full design rationale see docs/ARCHITECTURE.md.
| Layer | Choice |
|---|---|
| Language | Python 3.12+ |
| Bot framework | aiogram 3.x (FSM, routers, middlewares) |
| LLM | Anthropic / OpenAI (pluggable, OpenAI-compatible endpoints) |
| Market data | BingX USDT-M perpetual swap public API |
| Payments | Crypto Pay API (@CryptoBot) |
| Storage | SQLite (aiosqlite, WAL, file-based migrations) |
| Scheduling | APScheduler |
| Config | pydantic-settings (.env) |
| HTTP | httpx (async) |
| Deploy | Docker + docker-compose |
app/
main.py — entrypoint (long polling + APScheduler)
bot.py — Bot/Dispatcher factory, middlewares, routers
config.py — Settings (pydantic-settings)
handlers/ — start · subscription · analysis (FSM) · admin · errors
middlewares/ — db session · subscription gate · throttling
services/ — analysis pipeline · BingX · indicators · cryptopay · editable …
ai/ — provider ABC, factory, Anthropic & OpenAI implementations
db/ — aiosqlite + repositories + SQL migrations
scheduler/ — background jobs
utils/ — texts · keyboards · formatting
tests/ — indicator unit tests
A more detailed map lives in docs/ARCHITECTURE.md.
- Python 3.12+
- A Telegram bot token from @BotFather
- A Crypto Pay token from @CryptoBot → Crypto Pay → Create App
- An API key for your chosen LLM provider (Anthropic or OpenAI)
git clone <your-repo-url> nexisq
cd nexisq
python -m venv .venv
# Windows: .venv\Scripts\activate
# Linux / macOS: source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env # Windows: copy .env.example .env
# edit .env — set BOT_TOKEN, ADMIN_IDS, CRYPTOPAY_TOKEN and your AI key
python -m app.mainIn Telegram: send /start, then /grant (admin-only, gives yourself a
subscription for testing), then send a trade idea as text.
docker compose up -d --builddocker-compose.yml reads the same .env and persists the SQLite database to a
./data volume.
Configuration is read once from .env (.env.example documents
every option). The essentials:
| Variable | Required | Description |
|---|---|---|
BOT_TOKEN |
✅ | Telegram bot token |
ADMIN_IDS |
✅ | Comma-separated admin Telegram IDs (111,222) |
CRYPTOPAY_TOKEN |
✅ | Crypto Pay API token |
AI_PROVIDER |
✅ | anthropic or openai |
ANTHROPIC_API_KEY / OPENAI_API_KEY |
✅¹ | Key for the selected provider |
OPENAI_BASE_URL |
Custom OpenAI-compatible endpoint (e.g. OpenRouter) | |
SUB_PRICE / SUB_ASSET / SUB_DURATION_DAYS |
Subscription defaults (editable later in-bot) | |
BINGX_API_KEY / BINGX_API_SECRET |
Optional — BingX market data is public |
¹ One key, matching AI_PROVIDER.
Subscription price/asset/duration, AI model names, all user-facing texts and the AI instruction itself are editable at runtime via
/texts— no restart required. The.envvalues are just the starting defaults.
Crypto Pay: the token and
CRYPTOPAY_BASE_URLmust be on the same network. For testing usehttps://testnet-pay.crypt.bot/api/with a token from @CryptoTestnetBot.
For users
| Command | Action |
|---|---|
/start |
Intro, welcome animation, subscribe / greeting |
| (text) | Send a trade idea in plain language to get it analysed |
For admins
| Command | Action |
|---|---|
/stats |
Users / subscriptions / revenue / analyses (all-time, today, 7d) |
/grant [days] |
Give yourself a subscription without paying (testing) |
/revoke |
Remove your own subscription (to test the paywall) |
/texts |
Full editor for messages, the AI instruction and parameters |
pip install -r requirements-dev.txt
pytest -qThe suite covers the pure indicator functions (SMA, EMA, RSI, ATR, swing levels, volume trend, percent change) and runs without any network access.
NexisQ is an analysis tool, not financial advice. Crypto-perp trading carries substantial risk of loss. The bot evaluates the execution quality of an idea you provide — it does not predict prices and does not guarantee outcomes. Every decision is yours.
Released under the MIT License.