| 🎓 **Numex Okul** | Students, parents, teachers | One AI writes each question, a second AI solves it independently; mismatches are discarded. LGS/TYT/AYT exam prep, driving test; solve on screen or export PDF | [okul.numexai.com.tr](https://okul.numexai.com.tr) |
| 🧩 **Numex Codex** | Developers | Turkish coding agent: browser editor, desktop IDE (v2.9.59, VS Code based), VS Code extension. Modes: Chat · Plan · Autonomous · Bug Hunter | [codex.numexai.com.tr](https://codex.numexai.com.tr) |
| 🛍️ **Numex Market** | Everyone | 44 free, open-source apps Numex built autonomously, in 7 categories — try live, read the source, fork | [market.numexai.com.tr](https://market.numexai.com.tr) |
| ⌨️ **Numex CLI** | Developers | Autonomous agent in the terminal (v3.3.10): plan mode, long-running missions, undo, offline via Ollama, remote-PC bridge, MCP | `npm i -g @numexai/cli` |
| 🧰 **Codex SDK** | Developers | chat · embeddings · images · models · search + a CLI Bridge to embed the agent anywhere | `npm i numexcodex-sdk` |
| 🔌 **Developer API** | Teams | REST v1 with `numex-pro`, `numex-fast`, `numex-think`, `numex-vision`, `numex-code`; streaming, function calling | [API docs](urunler/05-api-ve-sdk.md) |
| 📦 **Numex Hub** | Developers | AI that knows your repo: ask for a change, review it, apply → committed. Community showcase, trending repos, live feed | [hub.numexai.com.tr](https://hub.numexai.com.tr) |
| 🏗️ **Numex Forge** | Developers | Gitea-based Git server with a fully Turkish UI | [forge.numexai.com.tr](https://forge.numexai.com.tr) |
| 📖 **Numexpedia** | Everyone | Computing, AI and software explained in plain Turkish — free and open | [pedia.numexai.com.tr](https://pedia.numexai.com.tr) |
| 🧭 **Pusulam** | SMEs | Bookkeeping & point of sale | [pusulamx.com.tr](https://pusulamx.com.tr) |
| 🩺 **PC Doktoru** | Windows users | Portable AI maintenance, diagnosis and repair with autonomous mode and rollback | [pcdoktoru.com.tr](https://pcdoktoru.com.tr) |
## ⚡ What makes Numex different
**Not a raw API wrapper.** Numex wraps leading model providers with its own engines:
1.**Pipeline** — every request goes through 5 stages: prompt enrichment → model → Turkish correction
& tone → quality check → answer. Stage 2A alone improved Turkish output quality by 30% in v3.1.
2.**Detective Mode™** — for critical questions, several models solve the same problem independently
(fast, deep, alternative) and a referee picks the best answer *with its reasoning*.
*"One model can be wrong. Three at once won't be."*
3.**Multi-Agent** — an orchestrator splits complex requests into sub-tasks for specialist agents and
merges the result into one consistent answer.
4.**DeepView™** — multi-expert orchestration that produces a single *Master Plan* and shows how the
AI thinks across 6 layers.
5.**Numex Core** — the agent engine behind Codex and the CLI: a 4-agent **Swarm Council**
(Architect, Coder, Auditor, Designer), a **self-healing loop**, **Omni-Vision** (headless-browser
DOM/3D checks), 3-tier inference (local ONNX WASM → Ollama → cloud), and **FinishGate**: a task
cannot be marked "done" without live evidence — HTTP 200, passing tests, or DOM checks.
**Proof over claims** runs through the whole family: Detective Mode in chat, dual-AI verification
in Okul, FinishGate in code — and Numex Market, where 44 autonomously built, verified apps are open
for anyone to inspect.
## 🇹🇷 Cultural characters
Call them with `@`: **@FatmaAna** (cooking, home economics), **@MuhasebeciYunus** (tax, VAT,