The whole trading loop, run by the same agent
Most tools hand you a signal and walk away. Cortiq runs the entire cycle — and feeds every live session's memory back into the next decision.
Spin up a single-instrument workspace. The agent pulls D1→M15 candles from MT5 and runs a structured edge-discovery pass — train-only hypotheses, out-of-sample validation, a written report you can read.
Turn findings into explicit rules: setup, entry, risk, management and invalidation. Attach preparation packages, sentiment reports and an instrument profile so every session starts from real context.
Replay the playbook over historical M1 with real AI decision cycles — strictly no-lookahead, with as-of prep and sentiment. Measure expectancy before a cent of capital is at risk.
Promote the same loop to execution. The agent wakes on candle close, price or time, reads the market, then opens, modifies and manages trades on your MT5 — inside hard external guardrails.
Every trade is auto-journaled with the agent's reasoning, then folds back into research and the next session. The loop closes.
See a full session, start to finish
No narration over slides. The real app: connect MT5, build the playbook, backtest it, go live, then review — every AI call and every guardrail on screen.
Everything the agent needs — and everything you need to trust it
Prep, data, indicators, risk and the full trade record live as inspectable artifacts. The AI proposes; your rules decide what trades. Nothing happens that you can't read back later.
Open any trade and see exactly why it happened
Cortiq stitches the full record from MT5: the entry, stop and target, the agent's reasoning at the moment of the call, every modification, and the chart at exit. The journal is the source of truth — not a summary you have to take on faith.
The agent walks in prepared
Preparation packages (higher-timeframe regime, key levels, structure, directional skew, session map), sentiment reports and an instrument profile are auto-generated before the session and refreshed on a schedule.
You decide what it sees
Pick the timeframes (M1→D1), indicators (MACD, RSI, Bollinger, MA, ATR, Stochastic, CCI…) and news the agent reads each cycle. Bundle them into reusable packages the agent receives on every wake.
Read straight from MT5
Indicator values are read live from your MetaTrader 5 terminal each cycle — the same numbers your charts show, handed to the agent as structured data instead of a screenshot.
Limits live outside the model
Global and per-account validators enforce daily and weekly drawdown, profit targets, max trades, per-symbol caps and loss-streak stops. A breach triggers an emergency stop that cancels pending orders across every account and pauses sessions — then auto-resumes when limits clear.
Every trade reconstructed
Entry, stop, target, the agent’s reasoning, the action timeline and the chart at exit — stitched together from MT5. Tag and annotate, generate session summaries, and tie into Open Market Journal.
Edge discovery, written down
Single-instrument workspaces run a three-phase flow — train-only hypotheses, out-of-sample validation, a playbook draft — and write a readable markdown report alongside the candle data.
Real decisions, no lookahead
Replay a playbook over historical M1 with real AI decision cycles, strictly no-lookahead, with as-of preparation, sentiment and profile context and the same risk simulation. Pause and resume; trades, cycles and metrics persist.
Human in the loop
The agent’s proposed memory, trigger and trade changes surface as approval cards in a Copilot dock. Mutating actions stay gated until you approve, inside the session windows you set.
Paper-trade the exact loop
Run the identical loop with no real orders. When a virtual trade earns it, promote it to a real MT5 order — same logic, same guardrails, one click.
Plug in any model. We never sit between you and your AI.
You choose the provider — your keys, your cost, your data. The agent auto-routes to whichever transport you've configured, and you can switch providers without losing the session. With a local model your ongoing AI cost can be €0.
One key, hundreds of models. Route to whatever is best for the job and change your mind whenever you like.
Ollama, LM Studio or llama.cpp over an OpenAI-compatible endpoint — fully private on your own GPU (a Qwen 3 model fits a 12GB card). Nothing leaves your machine.
Drive the agent through your existing ChatGPT or Claude subscription via the official CLI / agent — no extra API bill on top of the plan you already pay for.
Paste a key straight from OpenAI, Anthropic or Gemini. Billed by your provider, at their rate, with no middleman.
Built like infrastructure, not a bot.
Cortiq wires the agent into MetaTrader 5 through a local bridge, then wraps every action in hard limits that live outside the model. The reasoning is the AI's; the rules are yours.
The agent talks to MetaTrader 5 over a bridge that runs on your own machine — no broker plugin, no cloud relay between you and your order.
Run multiple MT5 terminals and accounts simultaneously, each on its own port. Every session binds to one chosen account.
Market and pending orders (limit/stop), SL/TP modification, partial close, full close and cancel — with a pre-trade broker margin check before anything is sent.
By design, each session is locked to a single market. No basket scanning — the agent builds deep, measurable expertise on the instrument in front of it.
Global and per-account rules enforce daily and weekly drawdown, profit targets, max trades, per-symbol caps and loss-streak stops — checked outside the model on every cycle.
A breach cancels pending orders across every account and pauses sessions, then auto-resumes once the limits clear. Unrealized P/L is monitored continuously.
Position size is computed from equity × your risk% and overrides whatever lot the agent proposed. The model can suggest; it cannot oversize.
A human-in-the-loop Copilot dock gates mutating actions until you approve, with time-of-day session windows that keep the agent inside the hours you trade.
The AI proposes. Your rules dispose.
The journal writes itself, then teaches the next session
Cortiq reconstructs every trade with the agent's reasoning and a full timeline, then folds it back into the loop. It's the same engine behind Open Market Journal, our public trade journal.
Entry, stop, target, the agent’s reasoning and the full timeline are reconstructed the moment a trade closes — no separate write-up.
The why behind each decision is captured as readable text you can re-read, tag and annotate. Nothing about the trade is hidden.
Each session’s memory feeds back into research and tomorrow’s prep, so the agent compounds what it learns instead of starting cold.
The questions a skeptic asks first
You should be wary of handing an agent real money. So here is exactly where the control sits — and where it doesn't.
No. Every AI call is plain text you can read, edit and reject before it acts. The agent’s reasoning, the market read and each decision are logged in full — nothing executes that you can’t inspect, before, during and after the trade.
Good. The risk limits live outside the model and the model cannot override them — position size, drawdown and trade caps are enforced by validators, not the AI. Mutating actions wait for your approval, and you can paper-trade the exact loop first.
Usually because they scan everything and size blind. Cortiq stays on one market, enforces hard daily and weekly drawdown stops, and gives you an emergency cancel-all across every account. You set the rules; a breach pauses sessions on its own.
It is a guided Windows install and an activation key you paste on first launch. Connect your MT5 account, point it at an AI you already have — OpenRouter, a local model, or your existing ChatGPT or Claude — and run.
Depth over breadth. Locked to a single instrument, the agent builds measurable expertise — its prep, profile and journal all compound on the same market. Multi-symbol scanning is left out on purpose, not missing.
It will be, sometimes — it is a tool, not a promise. That is why your playbook carries explicit invalidation rules, the guardrails cut losers on their own, and you stay in the loop on every mutating action. Cortiq does not predict markets or offer signals.
Early traders, real sessions
“The difference is that I stop second-guessing the plan mid-session. The playbook is already written — I just run it.”
“I’ve kept journals in Notion for years. This is the first time review actually loops back into tomorrow’s prep without me rewriting anything.”
“It’s useful precisely because it doesn’t try to be smart. The guardrails are boring, strict, and completely outside the AI.”
One key. One price. Yours to keep.
Cortiq is a one-time purchase — not a subscription, no auto-renewal. The only ongoing cost is your own AI usage, and with a local model that can be €0.
Cortiq is an execution and research environment, not a signals service and not financial advice. Trading is risky and you can lose money. Read the refund policy and risk disclosure before you buy.
Straight answers before you buy
What Cortiq is, what it needs, and what it deliberately does not do. No overpromises.