0mni

A TUI coding agent that combines Claude, Gemini, and Codex into a single CLI. Uses your locally installed coding agents as the runtime — no API keys needed.

┌─────────────────────────────────────────┐
│ 0mni v0.1.0 · ● claude-sonnet-4-6      │
├─────────────────────────────────────────┤
│                                         │
│ You: Fix the login bug in auth.ts       │
│                                         │
│ ● Claude: I'll look at auth.ts...       │
│ → read_file src/auth.ts                 │
│ → edit_file src/auth.ts                 │
│                                         │
│ Done. Fixed the null check on line 42.  │
│                                         │
├─────────────────────────────────────────┤
│ > _                                     │
└─────────────────────────────────────────┘

Install

npm install -g omni-cli

Or run directly:

npx omni-cli "fix the failing tests"

Prerequisites

You need at least one of these coding agents installed and authenticated:

• Claude Code — npm install -g @anthropic-ai/claude-code
• Gemini CLI — npm install -g @google/gemini-cli
• Codex CLI — npm install -g @openai/codex

0mni uses your existing CLI authentication. No separate API keys required.

Usage

# Interactive TUI
omni

# Start with a prompt
omni "explain this codebase"

# Force a specific provider
omni --provider claude "fix the bug"
omni -p gemini "search for the error"
omni -p codex "refactor the auth module"

# Auto-route to best provider per task
omni --auto "find and fix the performance issue"

# Specify a model
omni -m claude-opus-4-6 "design the new API"

# Auto-approve all tool calls
omni --yolo "fix all lint errors"

Slash Commands

Inside the TUI, use these commands:

| Command | Description | |---------|-------------| | /claude | Switch to Claude | | /gemini | Switch to Gemini | | /codex | Switch to Codex | | /auto | Toggle auto-routing | | /model <name> | Set a specific model | | /clear | Clear conversation | | /cost | Show token usage and cost | | /help | Show all commands |

Auto-Routing

When --auto is enabled, omni classifies your task and routes to the best provider:

| Task Type | Provider | Why | |-----------|----------|-----| | Complex reasoning, debugging | Claude | Best at multi-step reasoning | | Code generation, review | Claude | Strong code understanding | | Search, web queries | Gemini | Built-in Google Search grounding | | Multimodal (images) | Gemini | Best multimodal support | | Refactoring | Codex | Optimized for code transforms | | Simple Q&A | Claude Haiku | Fast and cheap |

How It Works

0mni is a TUI wrapper that spawns your installed coding agents as subprocesses:

0mni TUI (React + Ink)
    │
    ├── claude -p "prompt" --output-format stream-json
    ├── gemini -p "prompt" --output-format stream-json
    └── codex exec "prompt" --json

Each CLI handles its own:

• Tool execution (file editing, shell commands, search)
• Authentication (OAuth, API keys)
• Sandboxing and permissions

omni parses the streaming JSON output and displays it in a unified terminal UI.

Skills

0mni supports skills.sh for extensibility:

# Install skills
omni skills add vercel-labs/agent-skills
omni skills add anthropic/skills

# List installed skills
omni skills list

Configuration

Create .omni/config.toml in your project or ~/.config/omni/config.toml globally:

defaultProvider = "claude"
autoRoute = false
yolo = false

[models]
claude = "claude-sonnet-4-6"
gemini = "gemini-2.5-flash"
codex = "gpt-5.4"

License

MIT