AI integrations

GitWand can connect to an AI model to generate commit messages, PR titles and descriptions, branch names, stash messages, release notes, and conflict explanations. The model runs through a provider you configure — GitWand never bundles its own model.

Go to Settings → AI to pick a provider and configure it.

---

Claude Code CLI (recommended)

If you already use Claude Code, this is the simplest option. GitWand calls the local claude CLI subprocess and piggybacks on your existing Max or Pro subscription — no API key, no extra cost.

Requirements: Claude Code installed and logged in (claude login).

Setup:

1. Open Settings → AI
2. Select Claude Code CLI as the provider
3. That's it — GitWand detects the claude binary automatically

This is the same Claude Code you use in your terminal for coding tasks. GitWand uses it for focused, single-turn prompts (commit message from a diff, PR description from a branch, etc.) — it won't start interactive sessions or consume your conversation history.

---

Anthropic API

A direct connection to the Anthropic API with your own key.

Setup:

1. Open Settings → AI
2. Select Claude API
3. Paste your API key (from console.anthropic.com)
4. Choose a model — claude-sonnet-4-5 is a good default for speed and quality

Costs are billed to your Anthropic account per token. Commit message generation typically uses a few hundred tokens per call.

---

OpenAI-compatible endpoint

Any API that speaks the OpenAI chat completions format: OpenAI directly (gpt-4o, o3…), Azure OpenAI, OpenRouter, or a self-hosted endpoint.

Setup:

1. Open Settings → AI
2. Select OpenAI-compatible
3. Enter the endpoint URL — for OpenAI: https://api.openai.com/v1
4. Paste your API key
5. Enter the model name (gpt-4o, o3-mini, etc.)

---

Ollama (local, offline)

Run a model entirely on your machine with Ollama. No API key, no data leaves your computer.

Setup:

1. Install Ollama and pull a model: ollama pull llama3
2. Make sure Ollama is running (ollama serve)
3. Open Settings → AI
4. Select Ollama
5. Set the endpoint to http://localhost:11434 (default)
6. Enter the model name (llama3, mistral, codellama, etc.)

Good models for code-related tasks: codellama, deepseek-coder, qwen2.5-coder.

---

What the AI is used for

Once a provider is configured, AI features become available throughout the app:

Feature	Where
Commit message generation	Commit panel → ✦ button
PR title & description	PR Create view → ✦ button
Branch name suggestion	Branch creation dialog → ✦ button
Stash message	Stash panel → ✦ button
Release notes	Tags panel → ✦ button
Conflict explanation	Merge editor → Explain button
Conflict resolution suggestion	Merge editor → IA button
Per-hunk code review	PR detail → Intelligence panel
Semantic squash message	Interactive rebase → squash action
Blame context	File blame → ✦ button
Absorb ranking	Absorb flow (semantic ranking)
Merge risk summary	Merge preview panel

Every AI action is explicit — GitWand never calls the model in the background without your input.

---

Privacy

GitWand sends only what's needed for each task:

• Commit message: the staged diff
• PR description: the branch diff against base
• Conflict explanation: the ours/theirs/base content of the hunk

No file paths outside the diff, no repo history, no credentials. If you use Claude Code CLI or Ollama, nothing leaves your machine.