OMK CLI
Open-source multi-agent coding CLI: switch models mid-session and finish a goal only when your own test passes
- 152 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The agent runs commands in the repository and edits files, and you approve the check command by hand
- Login uses API keys and subscriptions stored locally
Install
Manual install
npm install -g open-multi-agent-kit --ignore-scriptsThen cd into the project and run omk. Without a global install: npx --ignore-scripts open-multi-agent-kit. A standalone binary installs with install.sh from the latest release.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent runs commands in the repository and edits files, and you approve the check command by hand
- Login uses API keys and subscriptions stored locally
This is third-party code. Review the repository files before installing.
What it does
OMK (Open Multi-Agent Kit) is a terminal coding agent written in TypeScript. Its main idea is a check in place of claims: the /goal verify command names your check, for example npm test. After each settled turn OMK asks for your approval of the command, runs it and repeats, and a goal completes only when the check passes on the current workspace. Any later edit makes earlier evidence stale. In one session you can switch models with /model and come back to the conversation through /resume or omk -c. Login works by subscription, API key or local model, and an existing Claude Code or Codex CLI login is reused with omk provider adopt. Built-in bash runs in a sandbox-exec sandbox on macOS or bwrap on Linux with no network and stops if the mechanism is missing. omk doctor checks the environment and prints fixes. By default one agent and tool loop runs. Subagents, MCP servers, extra skills and extensions load only when you configure them. OMK runs as a standalone program and is not a plugin for Claude Code or OpenCode.
Who it is for. Developers who want a terminal agent with a verifiable definition of done and freedom to switch models.
Good fit when
- A failing test must be fixed so that the goal closes only after a green check
- Midway through a session you want to switch to another model and ask it to review the answer
- Work on a task has to pause and resume later with `omk -c`
- You want to reuse your Claude Code or Codex CLI login without signing in again
Not a fit when
- You need a plugin inside Claude Code or OpenCode: OMK installs as a separate program
- There is no test command or git repository: a verified goal rests on a command that exits 0
- You need several models running in parallel: switching in a session is manual
Example request
Fix the failing test without changing its assertions and close the goal only after npm test passesLimitations
Node.js 22.19 or newer is required, or a standalone binary for macOS and Linux, with Windows builds attached to releases. Built-in bash needs sandbox-exec on macOS and bwrap with unprivileged user namespaces on Linux (the bubblewrap package on Debian and Ubuntu). A goal result is only as reliable as the check you choose. The --ignore-scripts flag in the install command disables npm package scripts.
How to disable. Remove the global package with npm uninstall -g open-multi-agent-kit, or the ~/.omk/bin directory for the standalone binary. Saved sessions are stored locally.
FAQ
How does /goal verify work?
You name a check command such as npm test. OMK asks you to approve it, runs it after each settled turn and closes the goal only on exit code 0 on the current workspace. Ordinary prompts without /goal verify do not enable the gate.
Which models are supported?
Subscriptions, API keys and local models, with the provider list in packages/coding-agent/docs/providers.md. Switch between models with /model without leaving the session.
What runs by default?
One agent and tool loop after provider setup: file editing, shell commands and saved sessions. Subagents, MCP, skills and extensions need your configuration.
How do I check that the environment is ready?
omk doctor checks the runtime, credentials, the model a first session would pick and the sandbox, then prints a fix for your machine.
How does OMK differ from Claude Code?
OMK installs as a separate CLI on Node.js or a standalone binary and switches models inside a session. A /goal verify goal finishes on a command check and not on the model's own judgment.
Discussion
No questions or reviews yet.
Related
CLI toolsKode CLI is a terminal coding agent that reads AGENTS.md and CLAUDE.md, connects MCP servers and works with many models.
A terminal coding agent in Go: GLM, DeepSeek and local Ollama models, 59 built-in tools and MCP
A CLI that runs Claude Code, Codex, Gemini CLI and other agents in parallel on one task and collects their answers side by side
A Python terminal coding agent in the manner of Claude Code: reads the repository, edits files, runs commands, with memory, sub-agents and MCP
A desktop board turns a card into an agent session in its own git worktree, flags the agent waiting on you, offers an MCP server
A small terminal coding agent in Rust: bring your own API key or local model, per-turn cost, real undo and session branches
Comparison with similar tools
Cards that share the main search query, and how they differ.
README in short
The README shows a one-minute start (npm, enter the project, run and /goal verify), explains why an agent saying done is not an acceptance check and lists what is on by default and opt-in. It continues with the binary install, the bash sandbox, context and memory, and evidence and limits.