Authentication
Pantoum uses Claude for AI-assisted recovery in the public release.
Choose One Authentication Method
You only need one of these:
Claude Code Subscription
If you already use Claude Code, Pantoum can reuse that authentication automatically. This is the easiest option.
Anthropic API Key
Alternatively, set the ANTHROPIC_API_KEY environment variable:
export ANTHROPIC_API_KEY=sk-ant-xxx
Get your API key from the Anthropic Console.
Verify Your Setup
Run the built-in doctor command:
npm run doctor
This checks Node.js, M365 CLI, Claude access, and the local Pantoum setup.
M365 CLI runs locally to generate upgrade reports. No Microsoft 365 tenant login is required.
Choosing A Model
The public release supports these Claude model values:
| Model | Flag Value | Best For |
|---|---|---|
| Sonnet (default) | sonnet | Most upgrades |
| Opus | opus | More difficult upgrades |
Set it via CLI flag or Advanced settings in Studio:
pantoum --localPath ./project --toVersion 1.23.0 --agentModel opus
GitHub Copilot
Pantoum can also run its AI stages on GitHub Copilot. Set agent_provider: "github-copilot" (or pass --agentProvider github-copilot) and pick a model with agent_model:
| Model | Flag Value |
|---|---|
| GPT-5 (default) | gpt-5 |
| GPT-5 mini | gpt-5-mini |
| MAI Code Flash (current variant) | mai-code-1-flash-picker |
| MAI Code 1.1 Flash (pinned) | mai-code-1.1-flash |
Authentication comes from the Copilot CLI's own login — install the GitHub Copilot CLI and sign in once with copilot, then Pantoum uses that session.
Model availability depends on your Copilot plan and, on Business/Enterprise plans, on the organization's Copilot model policies — some models are off by default until an administrator enables them. A model your plan does not serve fails at session start with Model "…" is not available; Pantoum reports this as a provider-side failure. mai-code-1-flash-picker routes to the MAI Code variant your plan currently serves, so it keeps working when pinned model ids are rotated or policy-gated — prefer it unless you need a specific pinned variant.
The prompts are the same for both runtimes. Only the execution backend changes.
Usage Visibility
Pantoum writes AI usage details into its reports so you can see what the AI did and how much it cost.