che-cli

A small, growing collection of personal command-line utilities — each one a subcommand of a single che binary.

The first tool, che commit, automates the boring git add . && git commit -m "..." loop by asking an LLM provider to summarize the diff into a commit message. The default provider is a local Ollama server — no cloud calls, no API keys, no telemetry. OpenAI and Anthropic are also supported.

$ che commit --push

→ refactor: extract ollama client into lib/che/llm.sh

commit with this message? [Y/n/e=edit] y
[main 3f1a2b8] refactor: extract ollama client into lib/che/llm.sh

Docs: chevp.github.io/che-cli Local LLM setup: cura-llm-local

---

Install

The installers do more than copy files: they detect missing dependencies (git, curl, python3, PyYAML, ollama, the default model) and install them through the platform's native package manager. Each install is confirmed interactively unless you pass --yes / -AssumeYes.

Common flag	Effect
--yes / -AssumeYes	say yes to every "install X?" prompt (unattended)
--no-deps	skip OS-package installs (don't touch git/python/ollama)
--no-ollama	leave ollama alone (don't install / serve / pull)
--no-model	install ollama but skip the multi-GB model pull

Linux / macOS / WSL

git clone https://github.com/chevp/che-cli.git
cd che-cli
./install.sh                  # interactive — confirms each install
./install.sh --yes            # unattended (CI-friendly)
PREFIX=/usr/local ./install.sh

This installs:

• ~/.local/bin/che — the dispatcher
• ~/.local/lib/che/ — full subcommand tree
• (if missing & confirmed) git, curl, python3, PyYAML, ollama, gh
• starts ollama serve in the background and pulls $CHE_OLLAMA_MODEL (default llama3.2)

Package-manager support: brew (macOS), apt-get, dnf/yum, pacman, zypper, apk. If ~/.local/bin is not already on your PATH, the installer appends an export PATH=... line to your shell rc (~/.zshrc, ~/.bash_profile/~/.bashrc, or ~/.config/fish/config.fish) — open a new terminal afterwards. Pass --no-path-edit (or CHE_NO_PATH_EDIT=1) to opt out.

Windows (PowerShell)

git clone https://github.com/chevp/che-cli.git
cd che-cli
.\install.ps1                 # interactive
.\install.ps1 -AssumeYes      # unattended
.\install.ps1 -Prefix "C:\Program Files\che"

This installs to %LOCALAPPDATA%\che and adds it to your user PATH. When dependencies are missing, it uses winget to install Git for Windows, Python 3, and Ollama (with a direct OllamaSetup.exe download as fallback), then installs PyYAML via pip --user and pulls the default model. Restart your terminal afterwards, then run che doctor.

Windows (Inno Setup installer)

A proper Windows installer is also available — it bundles the same dependency bootstrapper and runs it post-install.

cd installer
.\build.ps1

Output lands in installer/Output/. Double-click the .exe. Two opt-in tasks during the wizard:

• Install missing dependencies (Git, Python, Ollama) — checked by default
• Also pull the default Ollama model — unchecked by default (multi-GB)

Build prerequisites: Inno Setup 6.

Manual prerequisites (only if you opt out of dependency install): bash, git, curl, python3 (or python), and ollama for the default provider.

---

Commands

che commit — AI-summarized commit

Stages all changes, sends the diff to the configured LLM provider, gets back a one-line commit message, asks for confirmation, commits, and (optionally) pushes.

che commit              # stage + summarize + confirm + commit
che commit --push       # ... and push afterwards
che commit --dry-run    # just print the suggested message
che commit --edit       # open $EDITOR with the message pre-filled
che commit --yes        # skip the confirmation prompt

If git add emits warnings (e.g. the classic LF will be replaced by CRLF on Windows), they are forwarded to the active LLM provider, which proposes a single safe git config command to fix the root cause. The suggestion is shown with diagnosis + reasoning and only applied after you confirm. Skip this with CHE_FIX_WARNINGS=0.

che ship — recursive add + commit + push

For the current repo and every submodule (depth-first, recursively):

1. Initializes any uninitialized submodules (submodule update --init).
2. Fast-forward pulls if HEAD is on a branch (skipped if detached).
3. Runs che commit --push --yes — stage, AI-message, commit, push.

Submodules are processed before the parent so the parent's pointer-update commit references already-pushed children.

che ship                # one-shot recursive add + commit + push

Provider is selected via CHE_PROVIDER:

CHE_PROVIDER=ollama       che commit        # default — local llama3.2
CHE_PROVIDER=claude-code  che commit        # delegates to the `claude` CLI
CHE_PROVIDER=copilot      che commit        # delegates to the `copilot` CLI (GitHub)

Cloud LLMs are only reachable through their official CLIs (Claude Code's claude binary, GitHub's copilot binary). che never handles API keys directly — auth is owned by the CLI you've installed.

Configuration (all environment variables, all optional):

Variable	Default
CHE_PROVIDER	ollama
CHE_OLLAMA_HOST	http://localhost:11434
CHE_OLLAMA_MODEL	llama3.2
CHE_MAX_DIFF_CHARS	8000
CHE_FORCE_CLAUDE_CODE	unset (set to 1 to always escalate)
CHE_FIX_WARNINGS	1 (set to 0 to skip the LLM warning fixer)

If you don't have a local LLM running yet, follow cura-llm-local — about five minutes.

che run / che workflow — local pipelines

Runs a YAML manifest of shell-script steps from .che/workflows/<name>.yml. The workflow file is a pointer — it declares order, args, and inputs but contains no inline bash, so your existing scripts/ stay executable on their own. Mirrors .github/workflows/ for local pipelines.

che workflow list                       # discover workflows in this repo
che workflow show <name>                # print parsed step plan
che run <name> --tag=v1.2.3             # execute (alias: che workflow run)
che run <name> --tag=v1.2.3 --dry-run   # plan only

A workflow can also declare a trigger: (string or list) that registers che <trigger> as a top-level shortcut, including over built-ins like che ship:

# .che/workflows/ship.yml
name: ship
trigger: ship                # `che ship` now runs this file
steps:
  - name: tests
    script: scripts/test.sh
  - name: built-in ship
    script: ~/.local/lib/che/git/ship.sh   # original behavior, still reachable

A copy-paste template lives at .che/workflows/example.yml. Full guide: chevp.github.io/che-cli/workflow.html.

Requires python3 plus PyYAML (pip install pyyaml). Run che doctor workflow to check.

che doctor — health check

Verifies dependencies and providers. Platform-aware install hints (different suggestions for macOS vs Windows vs Linux/WSL).

che doctor               # everything
che doctor provider      # only the currently active provider
che doctor ollama        # specific target
che doctor docker        # ditto

Sample output:

platform: darwin
active provider: ollama (model: llama3.2)

git:
  ✓ git 2.39.5
docker:
  ✓ docker installed (29.3.1)
  ✓ docker daemon is running
ollama:
  ✓ ollama binary found
  ✓ server responding at http://localhost:11434
  ✓ model available: llama3.2

---

Project layout

che-cli/
├── bin/
│   └── che                       # dispatcher
├── client/http/                  # REST request samples (VS Code / JetBrains)
│   └── ollama.http
├── lib/che/
│   ├── platform.sh               # OS detection (darwin/windows/wsl/linux)
│   ├── provider.sh               # provider router (ollama/claude-code/copilot)
│   ├── doctor.sh                 # `che doctor`
│   ├── git/
│   │   ├── check.sh
│   │   └── commit.sh             # `che commit`
│   ├── ollama/
│   │   ├── check.sh
│   │   └── client.sh             # ollama_ping / ollama_generate / ollama_has_model
│   ├── claude-code/
│   │   ├── check.sh
│   │   └── client.sh             # wraps the `claude` CLI binary
│   ├── copilot/
│   │   ├── check.sh
│   │   └── client.sh             # wraps the `copilot` CLI binary
│   └── docker/
│       ├── check.sh
│       └── client.sh
├── docs/
│   └── index.md                  # GitHub Pages source
├── install.sh
└── README.md

Each provider folder exposes the same three functions (<p>_ping, <p>_has_model, <p>_generate) so provider.sh can route to whichever is selected by CHE_PROVIDER.

---

Adding a new tool

1. Drop a script at lib/che/<topic>/<name>.sh.
2. Add a case arm in bin/che that execs it.
3. Re-run ./install.sh.

No plugin manifest, no registration step — the dispatcher is short enough to read in one sitting (bin/che).

---

Related

• cura-llm-local — local Ollama setup guide
• chevp-mcp-suite — MCP server collection
• quon-cli — companion CLI tooling