Project-aware AI coding runtime

Install once. Resume every project. Route to the best coding model. Keep control of your code.

curl -fsSL https://get.omk.dev | sh

Inspect installer  ·  Security policy

See it in action Enterprise Relay Source on GitHub

30-second demo

OMK detects your project, mounts MCP servers, and routes to the best model — automatically.

$ omk ╭─ OMK — Project-aware AI coding runtime ─────────────╮ │ │ Project: ~/projects/my-app │ MCP servers: omk-project (16 tools) │ Skills: omk-context-broker, omk-plan-first │ Provider: DeepSeek Flash (auto-selected) │ ╰──────────────────────────────────────────────────────╯ you> add dark mode toggle to settings page ✓ Intent: code_edit | Risk: write | Model: deepseek-flash ✓ Plan created (3 steps) ✓ SettingsDarkToggle.tsx — implemented ✓ 4/4 tests passing ✓ Changes committed: feat(settings): add dark mode toggle you> /model kimi → Switched to managed:kimi-code for next turn you> /status Provider: managed:kimi-code | MCP: omk-project | Session: 12 min

Why OMK?

Not another AI wrapper. A runtime that remembers, routes, and verifies.

Memory

Project context

OMK remembers your project structure, conventions, and decisions across sessions.

Routing

Model lanes

Auto-route by task complexity. Fast models for simple edits, strong models for architecture.

MCP

Tool orchestration

Auto-connect project MCP servers. Skills, hooks, and evidence gates — one command.

Governance

Consent & DLP

Granular consent levels. Redaction preview. Never auto-opt-in for dataset sharing.

Model lanes

OMK is provider-neutral. Route to the best model for each task.

Default

DeepSeek Flash

Fast, cheap, high-volume coding runs. Great for routine edits and quick iterations.

Challenge

MiMo Pro

Long-horizon coding comparison. Test against frontier reasoning models.

Quality

DeepSeek Pro

Higher-quality verification baseline. Use for architecture and review tasks.

Premium

Kimi / Claude / Codex

Benchmark and comparison lane. Bring your own API key or use Relay credits.

Privacy & consent

Your code is yours. Dataset sharing is always opt-in, never automatic.

L0
Ops stats
Anonymous usage metrics
Selectable
L1
Trace meta
Intent, model, duration
OFF by default
L2
Trajectory
Tool calls, diffs, test results
OFF by default
L3
Sample pack
Redacted, per-case approval
Per-case
L4
Raw code
Source code content
Forbidden

무료 request 제공 ≠ 데이터 제공 동의. Free relay credits do not require dataset contribution. You can use OMK fully without sharing any data.

omk consent status —查看 current settings
omk consent preview — see what would be shared
omk consent enable l2 — opt in to trajectory sharing
omk consent revoke — revoke all consent

OMK Relay SaaS

Secure model routing, budget control, redaction, and audit logs for teams.

Free

Free Relay

Free coding requests with auto-selected models. No data sharing required.

Team

Team Workspace

Shared project memory, budget controls, team consent ledger.

Enterprise

Private Relay

Private routing, DLP/redaction, audit logs, deletion SLA. On-prem available.

Benchmark

Secure Benchmark Room

Isolated evaluation environment. Compare models on your code without exposure.

Join the beta

Get early access to OMK Relay, free credits, and the K-Coding Dataset consortium.

Security

Supply-chain trust is non-negotiable.

Install script: get.omk.dev/install.sh — inspect before running

Checksums: GitHub Releases — SHA-256 for all artifacts

Signatures: GitHub Releases — cosign / Sigstore

Security policy: SECURITY.md

The installer does not read API keys, .env files, or repository code. No telemetry is sent before first successful run.