Private beta — opening access Q2 2026

Your IDE.
Your model.
Your rules.

Hypex is an agentic IDE built on Code OSS. Bring any model — Claude, GPT-5, Qwen, DeepSeek, GLM — and ship real work without an AI assistant refusing your job.

Open source core Free in BYOK mode Windows, macOS, Linux
hypex driver.c agent
driver.c ioctl.h loader.asm
 1  #include <ntddk.h>
 2  #include <wdm.h>
 3  #include "driver.h"
 4  #include "ioctl.h"
 5
 6  // IOCTL dispatcher — handles IOCTL_GET_PROCESS_HANDLES
 7  NTSTATUS DispatchDeviceControl(
 8      PDEVICE_OBJECT DeviceObject,
 9      PIRP            Irp)
10  {
11      PIO_STACK_LOCATION io = IoGetCurrentIrpStackLocation(Irp);
12      ULONG code = io->Parameters.DeviceIoControl.IoControlCode;
13
14      switch (code) {
15          case IOCTL_GET_PROCESS_HANDLES:
16              return HandleGetProcesses(Irp);
17          default:
18              return STATUS_INVALID_DEVICE_REQUEST;
19      }
20  }
21
22  NTSTATUS DriverEntry(PDRIVER_OBJECT drv, PUNICODE_STRING reg)
23  {
24      UNICODE_STRING name = RTL_CONSTANT_STRING(L"\\Device\\Hypex");
25      return IoCreateDevice(drv, 0, &name, FILE_DEVICE_UNKNOWN, ...);
26  }
Ln 16, Col 42 UTF-8 C agent ready
Claude Sonnet · agent 4.5
you
Add an IOCTL handler that returns active process handles. Wire it to PsLookupProcessByProcessId.
agent
Added IOCTL_GET_PROCESS_HANDLES handler, wired to PsLookupProcessByProcessId. Run msbuild to verify.
+ src/driver.c +42 −3
+ src/ioctl.h +8 −0
ask anything ⌘K
/ download

Get Hypex. Today.

Private beta build. Unsigned for now — SmartScreen will warn on first launch. Click "More info" then "Run anyway". Signing lands in v0.2.

Windows 10 / 11

Download Hypex for Windows

User-level installer. No admin rights needed. Installs to %LOCALAPPDATA%\Programs\Hypex.

Download .exe x64 ~180 MB Free · BYOK
macOS

Coming soon

Universal .dmg · Intel + Apple Silicon · v0.3

Linux

Coming soon

.deb · .rpm · .tar.gz · v0.3

SHA256
TBD — published with release notes
Build from source
/ features

An IDE that actually helps you ship.

Four things we refused to compromise on.

Developer-first, not nannied

The agent cooperates on real engineering work: reverse engineering, security research, scrapers, automation. No hand-wringing about legitimate technical tasks.

Example: ask Hypex to dump offsets from a binary you own — it does it.
Respects ToS. No malware. Just real work.

Multi-model BYOK

Plug any provider. Switch per task. Claude Opus for deep refactors, Qwen3-Coder for bulk edits, DeepSeek-V3 for cost. Your key, your bill, your choice.

Example: Claude Opus for architecture, swap to Qwen for bulk edits, DeepSeek for cost.
Anthropic · OpenAI · OpenRouter · Ollama

Free in BYOK, €15 for Pro

Core IDE and agent are free forever when you bring your own key. Pro unlocks cloud sync, the MCP marketplace, and priority support — no per-seat tax.

Example: indie devs pay €0. Teams that want sync pay €15/seat. That's the whole menu.
No surprise invoices. Ever.

Built for drivers and C++

First-class MSBuild, .vcxproj, WDK and kernel-mode support on the roadmap. The only agentic IDE that understands what a device stack is.

Example: edit a KMDF driver with an agent that actually knows IRQL and DDI rules.
Kernel, embedded, firmware — v1.0 target.
/ built for serious work

The kind of code that generic assistants refuse.

Hypex is shaped around the fields where AI tools routinely get in the way. If your daily work looks like this, you're home.

Kernel drivers

KMDF, WDM, IRQL-aware edits. The agent knows when to keep paged pool out of a DISPATCH_LEVEL path.

Security research

Write PoCs for CVEs, fuzz harnesses, exploit primitives — on binaries you own or have authorization for.

Game development

Unity, Unreal, Godot, shader code. Mod single-player games, build tools, automate pipelines.

Embedded C++

ARM Cortex, ESP32, STM32. Memory-constrained code review and peripheral-aware refactors.

Reverse engineering

Disassembly analysis, IDA/Ghidra output triage, protocol reconstruction. Clean technical work.

Data engineering

Airflow, dbt, Spark, DuckDB. Pipelines that actually run, not glossy notebooks.

Not in this list? If it's legitimate professional work, Hypex is for you too.

/ models

Choose your model. Switch any time.

Hypex is provider-neutral. Paste an API key, pick a model per task, or route everything through OpenRouter. Your keys stay local — they never touch our servers.

# ~/.hypex/config.toml
default_model = "claude-sonnet-4.5"
fallback = "qwen3-coder"
local_only = false
A
Anthropic
Claude Opus · Sonnet · Haiku

Best for deep refactors, reasoning, and long contexts.

OpenAI
GPT-5 · o-series

Frontier general-purpose reasoning and tool use.

OpenRouter
Qwen3-Coder · DeepSeek-V3 · GLM-4.6

Open-weight models at a fraction of the price. One key, 200+ models.

Ollama
Local · air-gapped

Run models on your own hardware. No data ever leaves your machine.

/ compare

The honest comparison.

No asterisks. Here's where Hypex sits next to the usual suspects.

Hypex This one.
Cursor
Kiro
VS Code
Starting price Free $20/mo Paid Free
Bring your own API key
Partial limited
Multi-model (Claude · GPT · OSS)
Partial via extensions
Permissive on legitimate technical work
Partial
WDK / kernel dev roadmap
roadmap v1.0
Partial
Proper agent framing (not prompt tricks)
Open source core
Local / air-gapped models
Partial
VS Code extensions
Partial

Last updated April 2026. We'll keep this table honest as the competition shifts.

/ pricing

Priced to not be annoying.

Free if you bring a key. €15/mo if you want the niceties. That's it.

Free

Available now
€0 / forever · BYOK

The full IDE and agent, free when you bring your own API key.

  • Full agentic IDE based on Code OSS
  • Bring your own key (Anthropic · OpenAI · OpenRouter · Ollama)
  • Unlimited local projects
  • All VS Code extensions
  • Community support
  • Early access to driver/WDK support
Get started free
Recommended

Pro

€15/mo — Coming soon
€15 / per month

Everything in Free, plus the team and power-user features.

  • Everything in Free
  • Encrypted cloud sync (settings, prompts, MCP servers)
  • MCP marketplace with curated servers
  • Team seats and shared contexts
  • Priority model routing & relay (optional)
  • Priority support + private Discord
Join Pro waitlist

BYOK means you pay your model provider directly. We never see your tokens.

/ roadmap

Where we're going.

A real roadmap. Dates will slip — we'll tell you when they do.

  1. MVP · Q2 2026

    Agent chat in the editor

    • — Fork of Code OSS with Hypex branding
    • — Agentic chat panel — multi-turn, tool use, file edits
    • — BYOK for Anthropic, OpenAI, OpenRouter, Ollama
  2. v0.5 · Q3 2026

    MSBuild & .vcxproj

    • — Native MSBuild task runner and diagnostics parsing
    • — .vcxproj understanding for the agent context
    • — MCP server marketplace (alpha)
  3. v1.0 · Q4 2026

    Driver & WDK

    — the killer feature
    • — WDK integration, kernel-mode debugging helpers
    • — PDB & symbol aware navigation for drivers
    • — Pro tier launch: cloud sync, teams, priority support
  4. v2.0 · 2027

    Embedded & firmware

    • — ARM Cortex, ESP32, STM32 first-class toolchain
    • — JTAG/SWD debugging via agent
    • — Offline air-gapped mode with local models
/ faq

Questions, answered straight.

No marketing, no lawyer-speak.

Is Hypex a fork of VS Code?
Hypex is built on Code OSS, the open-source base that underlies VS Code. You keep the editor you know, plus a first-party agent, rewind timeline, multi-model BYOK, and native build tooling for C/C++ and drivers.
Do I need to pay for AI models?
In BYOK mode, you pay your provider directly (Anthropic, OpenAI, OpenRouter, or run Ollama locally for free). Hypex takes zero cut on inference. A paid Pro tier adds cloud sync, premium MCP marketplace, and priority support — never model access.
Is Hypex permissive on technical work?
Yes, within every provider's Terms of Service. Hypex is developer-first: it won't refuse legitimate professional work like reverse engineering a binary you own, writing a security research PoC, building a scraper, automating a workflow, or modifying a single-player game. It still refuses malware targeting real victims, CSAM, weapons, and fraud. The difference vs generic assistants is framing, not rule-breaking.
When will Hypex be available?
Private beta Q2 2026, public launch Q3 2026. Join the waitlist to get early access and help shape the product.
Can I use my existing VS Code extensions?
Yes. Hypex is fully compatible with the VS Code extension API and the Open VSX marketplace.
Does it support driver and kernel development?
Roadmap v1.0. Native MSBuild and .vcxproj support, WDK templates (KMDF, UMDF), signtool integration, and a driver-aware agent that understands IRQL constraints and DDI rules. Today, Visual Studio 2022 is the only serious IDE for kernel work — and it has no modern AI agent. Hypex fills that gap.
What about privacy?
No content logging ever. Optional, opt-in telemetry captures only anonymous usage (model chosen, latency, error counts). Your code and prompts never touch Hypex servers — they go directly from your machine to the provider you picked.
Is my API key safe?
Keys are stored in your OS secret store via the VS Code SecretStorage API — Keychain on macOS, DPAPI on Windows, libsecret on Linux. They are never transmitted to Hypex servers.
Why BYOK instead of a flat-rate plan?
Cost control and no abuse risk. You pay exactly what you use, pick the cheapest model for routine edits and a frontier model for hard problems. A flat-rate plan forces pricing that assumes the worst-case user — BYOK doesn't.
What if my project contains sensitive code?
Hypex only sends the context you explicitly include. You control every file pin, steering file, and workspace indexing step. Nothing leaves your machine without a specific agent call that you can cancel or refuse.
Can I self-host everything?
Yes. Ollama integration is built-in from day one. Point Hypex at your local model, disable telemetry, and you have a fully offline AI IDE.
How is this different from tricks to bend the model?
Hypex doesn't try to bend anything. It uses proper system prompt framing and lets you pick a model whose defaults match your work. Everything stays within each provider's Terms of Service — cleaner, safer, and more reliable than prompt-level tricks.

Still have questions? Join the waitlist and ping us.

Beta opens Q2 2026

Be there on day one.

Get private beta access, a 14-day Pro trial at launch, and early influence on the roadmap.

No spam. We only email once the beta opens.

Already 42 developers waiting.

By submitting, you agree to the Terms and Privacy Policy.