@r3dlex/oh-my-gemini
v1.0.2
Published
Orchestration layer for Gemini CLI workflows
Readme
oh-my-gemini
Sister projects: oh-my-claudecode (OMC) | oh-my-codex (OMX)
Multi-agent orchestration for Gemini CLI with OMG branding and OMP compatibility aliases.
Transition status (2026-04-13): this repo now treats
oh-my-gemini/omgas the canonical user-facing surface, with legacyomp/.ompaliases retained only where compatibility still requires them. Seedocs/analysis/2026-04-13-oh-my-gemini-phase-1-doc-and-quality-review.mdfor the audited gap list and migration order.
Quick Start | Team Mode | Commands | Docs
Quick Start
npm install -g oh-my-gemini
omg setup --scope project
geminiAfter setup, restart Gemini CLI for /omg:* commands to appear (/omp:* remains compatible during migration).
The packaged extension now ships a Gemini-native hooks/hooks.json bridge and exposes omg_cli_tools as the canonical MCP server id.
omg doctor # check prerequisites
omg team run --task "..." --workers 2 # parallel work
omg hud --watch # live statusPrimary CLI launches (omg / omg launch) perform a throttled TTY-only update check (12h cache) and can prompt to run npm install -g oh-my-gemini@latest.
Disable the prompt with OMG_AUTO_UPDATE=0 (compatibility alias: OMP_AUTO_UPDATE=0).
Team Mode
tmux-first multi-worker orchestration with persistent state and lifecycle controls.
omg team run --task "review src/ for reliability gaps" --workers 4
omg team status --team oh-my-gemini --json
omg team resume --team oh-my-gemini
omg team shutdown --team oh-my-gemini --forceDefault backend: tmux | Optional: subagents for role-tagged runs
Commands
CLI
| Command | Description |
|---------|-------------|
| omg | Launch Gemini CLI with the oh-my-gemini extension |
| omg update | Update the globally installed package immediately |
| omg team run | Start orchestrated team run |
| omg team status/resume/shutdown/cancel | Team lifecycle |
| omg doctor | Diagnose prerequisites |
| omg verify | Run validation suites |
| omg hud | Live team status overlay |
| omg skill | List/print reusable skill prompts |
Slash Commands (inside Gemini CLI)
| Command | Description |
|---------|-------------|
| /omg:autopilot | End-to-end autonomous execution |
| /omg:plan | Phased execution plan with gates |
| /omg:execute | Immediate task implementation |
| /omg:review | Structured code review |
| /omg:verify | Acceptance validation |
| /omg:debug | Root cause investigation |
| /omg:status | Progress summary |
| /omg:cancel | Graceful stop |
| /omg:handoff | Context transfer document |
Full command reference: docs/omg/commands.md
Compatibility Note
User-facing command and documentation surfaces now use omg / oh-my-gemini; legacy omp / oh-my-product aliases are treated as compatibility-only shims.
Some internal compatibility identifiers intentionally remain unchanged for now:
- legacy hidden state and artifact paths
- legacy
OMP_*compatibility aliases for selected environment variables - legacy internal interop identifiers
- legacy internal type/class names
- the temporary
omp_cli_toolsMCP server alias (omg_cli_toolsis now the canonical id used by setup and extension manifests)
Those internal names are deferred to a later migration to avoid breaking state, protocol, and compatibility contracts.
Requirements
- Node.js 20+
- Gemini CLI
- tmux (
brew install tmux/apt install tmux)
Default Model
| Model | Free (OAuth) | Free (API Key) |
|-------|--------------|----------------|
| gemini-3.1-flash-lite-preview (default) | Yes | Yes |
| gemini-3.1-pro-preview (--pro) | Yes | Yes |
Default model on omg launch is gemini-3.1-flash-lite-preview. Use omg --pro for gemini-3.1-pro-preview (omp --pro also works).
All defaults work without a paid Gemini CLI coding plan — just log in via OAuth (Google Account) or use an API key. Override with OMP_MODEL_HIGH, OMP_MODEL_MEDIUM, OMP_MODEL_LOW env vars. Gemini 3.1, 3, and 2.5 models are all available via -m flag.
Emergency Model Override
If a default model becomes unavailable, override immediately without code changes:
export OMP_MODEL_HIGH=<working-model>
export OMP_MODEL_MEDIUM=<working-model>
export OMP_MODEL_LOW=<working-model>License
MIT
