A local-first CLI code agent powered by Gemma 4. Reads, writes, and refactors your codebase through natural language — with shell execution, git workflows, and project-aware context. Runs entirely on your machine via Ollama and OpenClaw.
No API keys. No cloud. No latency. Just nip it.
nip is an open-source terminal coding assistant that brings agentic AI to your local dev environment. Think Claude Code or Codex, but running Gemma 4 on your own hardware through Ollama. It understands your project structure, edits files surgically, runs commands, manages git, and indexes your codebase — all from a single REPL.
- Natural language coding — describe what you want, nip reads, writes, and refactors your files
- Local inference — Gemma 4 via Ollama, everything stays on your machine
- Tool calling — file I/O, shell execution, git operations, codebase search
- Project-aware context — auto-detects language, framework, and structure on startup
- Dual backend — direct Ollama (fast, simple) or OpenClaw gateway (MCP tools, skills, multi-channel)
- Safety built in — permission tiers, confirmation prompts for destructive operations, command blocklists
- Slash commands —
/diff,/undo,/compact,/modelto switch models mid-session
# Install
pip install nip
# Pull the model
ollama pull gemma4
# Run in your project directory
cd your-project
nipnip ready • gemma4 via ollama
Project: ./your-project
you > add error handling to the API routes
nip runs a ReAct agent loop — Gemma 4 reasons about your request, calls tools (read files, run commands, edit code, search symbols), observes the results, and iterates until the task is done. The full architecture and implementation details are in PLAN.md.
# Use a different model
nip --model gemma4:27b
# Connect through OpenClaw instead of direct Ollama
nip --backend openclaw
# Point to a specific project
nip --project /path/to/repoSee PLAN.md for the full implementation plan, phased rollout, and architecture diagrams.
MIT — see LICENSE for details.