What it does

OMK, short for Open Multi-Agent Kit, is a standalone command-line coding agent. One agent reads files, edits code and runs commands in your repository, and you can move to a different model with /model without leaving the session. Sessions can be saved and picked up again later with /resume or omk -c. Its distinctive feature is goal verification. With /goal verify you name a command; OMK runs it after each settled turn and marks the goal complete only when it passes on the current workspace, and a later edit makes that evidence stale. The site stresses that this is an explicit workflow, that ordinary prompts are not gated, and that the result is only as strong as the check you choose. Subagents, MCP servers, skills and extensions are optional additions rather than the default. It can sign in with a subscription, use API keys or run against local models, and the site lists providers such as Codex, Claude Code, OpenCode Zen/Go, Kimi, GLM/ZAI, xAI/Grok and NVIDIA NIM. Built-in local bash is sandboxed with sandbox-exec on macOS or bwrap on Linux and has network access blocked. The site is open about limits: its published Terminal-Bench 2.1 comparison against mini-SWE-agent found a success-rate difference that was not statistically significant, and it says state-of-the-art quality is not verified.

Capabilities

Switch models inside one session with /model
Durable goals with /goal verify that complete only when your chosen check passes
Resume saved sessions with /resume or omk -c
Sandboxed built-in bash on macOS and Linux with network access blocked
Optional subagents, MCP servers, skills and extensions
Works with subscription logins, API keys and local models

When to use it

You want an agent to count a task as done only when your own test command passes
You want to hand different steps of a coding task to different models
You prefer a terminal-based coding agent that can also run against local models