Omni

The Ultimate Git Experience in the Agentic Era

About

Omni is a copilot for git. It automatically enriches your commit history into a knowledge graph. Every git action you or your agent makes compounds that graph. The more you use it, the sharper your AI's reasoning becomes.

Why Omni

1. Omni runs in its own context window — a dedicated sub-agent — so your main agent never wastes tokens on git housekeeping and enrichment. Works across CLI, IDE, Agent, and CI/CD.
2. Omni outperforms any skill + MCP rig you can build yourself — backed by proprietary research and thousands of hours of A/B testing. Because enrichment compounds, your codebase improves automatically over time.
3. Prune your PRs: collapse 100s of commits into 10 razor-sharp commits, or split a messy refactor with 100s of file changes into clean, logical commits. (Git worktree support coming soon.)

Supported Operating Systems

• macOS Apple Silicon / Intel
• Linux x64 / ARM
• Windows x64 / ARM

Install

macOS / Linux:

curl -fsSL [https://edgeleap.github.io/omni.sh](https://edgeleap.github.io/omni.sh) | bash

Windows (PowerShell):

irm [https://edgeleap.github.io/omni.ps1](https://edgeleap.github.io/omni.ps1) | iex

CLI Commands

See CLI_Command_Reference.md for full command references.

Commit

• omni commit (quick commit)
• omni commit --preview (generate plan + print, do not apply)
• omni commit --edit (open plan file for editing / or launch editor)
• omni commit --apply (apply plan)

Flags:

• --json (agent output)
• --yes (CI/non-interactive apply)

PR

• omni pr --preview
• omni pr --edit
• omni pr --apply

Flags:

• --json (agent output)
• --yes (CI/non-interactive apply)

Release

• omni release --preview
• omni release --edit
• omni release --apply

Flags:

• --json (agent output)
• --yes (CI/non-interactive apply)
• --draft (GitHub release draft)
• --tag v1.2.3 (use explicit tag)

Agents

Omni integrates with coding agents at two levels: hooks (system-triggered, hard enforcement) and instructions (advisory rules the agent follows). Some agents support both; some support only instructions.

Agent	Hooks	Instructions	Enforcement	Integration
Claude Code	PreToolUse	CLAUDE.md	Hard block	Intercept script
Cursor	beforeShellExecution	.cursor/rules/*.mdc	Hard block	Intercept script
Cline	PreToolUse	.clinerules	Hard block	Intercept script
Gemini CLI	BeforeTool	GEMINI.md	Hard block	Intercept script
Copilot CLI	preToolUse	.github/copilot-instructions.md	Hard block	Intercept script
Copilot Agent	preToolUse	.github/copilot-instructions.md	Hard block	Intercept script
Codex CLI	—	AGENTS.md + execpolicy	Partial (sandbox + deny rules)	Policy rules
Codex App	—	Skills + AGENTS.md	Advisory (agent doesn't auto-commit)	Skill
Antigravity	—	Rules + Workflows + Skills	Advisory only	Rule + Workflow
Kimi CLI	—	AGENTS.md	Advisory only	Instruction file
Qwen Code	—	Subagents + project config	Advisory only	Subagent

How to pick your setup:

• Agent supports hooks? → Use the intercept script to hard-block git commit / gh pr create, plus an instruction file as backup. The agent literally cannot bypass omni.
• Agent has sandbox/policy rules? (Codex CLI) → Use execpolicy deny rules to block git commands, plus AGENTS.md to teach the agent what to use instead.
• Agent doesn't auto-commit? (Codex App) → No interception needed. Teach the agent about omni through a Skill, and it uses omni when you ask it to commit.
• Advisory only? (Antigravity, Kimi, Qwen) → Instruction files are your only option. The agent should follow them, but there's no system-level guarantee.

CI/CD

CI (two-step):

omni release --preview --tag v1.2.3 --json > /tmp/release-plan.json
omni release --apply --json

Note: GitHub release creation requires GITHUB_TOKEN/GH_TOKEN (GitHub Actions provides GITHUB_TOKEN automatically).