Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/jundot/omlx/llms.txt

Use this file to discover all available pages before exploring further.

The omlx launch command handles the full setup-and-start workflow for external coding tools: it writes the tool’s config file, sets the correct API endpoint and key, and then replaces the current process with the tool binary. All you need is a running oMLX server and the tool installed on your system.

​
How omlx launch works

When you run omlx launch <tool>, oMLX:
  1. Verifies the oMLX server is reachable at the configured host and port.
  2. Fetches available models from /v1/models.
  3. If --model is not specified, presents an interactive model picker (arrow keys, Enter to confirm).
  4. Writes or updates the tool’s config file with the oMLX endpoint, API key, and selected model.
  5. Exec’s the tool binary, replacing the current process.
# Interactive model selection
omlx launch codex

# Specify model directly
omlx launch codex --model Qwen3-Coder-Next-8bit

# Custom host and port
omlx launch codex --model Qwen3-Coder-Next-8bit --host 0.0.0.0 --port 8080

​
Codex

Codex integration writes ~/.codex/config.toml, adding oMLX as a named model provider and setting it as the active model. If the selected model name contains a reasoning hint (thinking, o1, o3, or r1), model_reasoning_effort = "high" is added automatically. 
omlx launch codex --model Qwen3-Coder-Next-8bit
The resulting config section looks like: 
model = "Qwen3-Coder-Next-8bit"
model_provider = "omlx"

[model_providers.omlx]
name = "oMLX"
base_url = "http://127.0.0.1:8000/v1"
env_key = "OMLX_API_KEY"
The OMLX_API_KEY environment variable is set to your oMLX API key (or "omlx" if no key is configured) before launching.
If Codex is not installed: npm install -g @openai/codex

​
OpenClaw

OpenClaw integration writes ~/.openclaw/openclaw.json with an omlx provider block and sets it as the default model. It also configures ~/.openclaw/exec-approvals.json based on the tools profile you choose. 
# Default tools profile (coding)
omlx launch openclaw --model Qwen3-Coder-Next-8bit

# Choose a different tools profile
omlx launch openclaw --model Qwen3-Coder-Next-8bit --tools-profile full

​
Tools profiles

The --tools-profile flag controls OpenClaw’s exec approval policy:
ProfileExec policyAsk behavior
minimalallowlistPrompt on miss
codingunrestrictedOff (default)
messagingallowlistPrompt on miss
fullunrestrictedOff
If OpenClaw has not been onboarded yet, omlx launch openclaw runs the non-interactive onboarding step automatically before writing the config.
If OpenClaw is not installed: npm install -g openclaw

​
Pi

Pi integration writes two files: ~/.pi/agent/models.json (provider and model definition) and ~/.pi/agent/settings.json (default provider and model selection). VLM models are configured with image input support automatically. 
omlx launch pi --model Qwen3-Coder-Next-8bit
Pi is launched with the model pre-selected: 
pi --model omlx/Qwen3-Coder-Next-8bit
If Pi is not installed: npm install -g @mariozechner/pi-coding-agent

​
OpenCode

OpenCode integration writes a provider entry to ~/.config/opencode/opencode.json using the @ai-sdk/openai-compatible npm package. The selected model is set as the default in the config, and VLM models are configured with image attachment support. 
omlx launch opencode --model Qwen3-Coder-Next-8bit
The resulting provider block: 
{
  "provider": {
    "omlx": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "oMLX",
      "options": {
        "baseURL": "http://127.0.0.1:8000/v1"
      },
      "models": {
        "Qwen3-Coder-Next-8bit": {
          "name": "Qwen3-Coder-Next-8bit",
          "modalities": { "input": ["text"], "output": ["text"] }
        }
      }
    }
  },
  "model": "omlx/Qwen3-Coder-Next-8bit"
}
If OpenCode is not installed: curl -fsSL https://opencode.ai/install | bash

​
Checking installed tools

Run omlx launch list to see every integration and whether the required binary is on your PATH: 
omlx launch list

Available integrations:
  claude       Claude Code (installed)
  codex        Codex (installed)
  openclaw     OpenClaw (not installed)
  pi           Pi (not installed)
  opencode     OpenCode (installed)

​
Admin dashboard alternative

All integrations are also accessible from the Integrations tab in the admin dashboard at http://localhost:8000/admin. The dashboard provides the same one-click setup without needing a terminal.

Build docs developers (and LLMs) love

Get started for freeTalk to us