First run

You need one thing before PP Agent can answer: a model. Everything else is optional and can be added later.

1. Connect a model

Open Settings › Models and either:

sign in to a built-in provider with an API key — Anthropic, OpenAI, Google, OpenRouter, Mistral, Groq, and more; or
add an OpenAI- or Anthropic-compatible gateway with a URL, key, and model names.

Keys are encrypted with a master key held in your operating system’s keychain. You can choose the model and thinking level per conversation; see Models and providers.

2. Add your business systems (optional)

Open Settings › MCP servers to connect tools such as a CRM, issue tracker, database, or internal API. You can add a server by hand, sign in with OAuth, or paste a configuration from Claude Desktop or Cursor. See MCP servers.

3. Create an agent

An agent is a profile for one kind of work: its own rules, tools, MCP servers, skills, model, starter prompts, and workspace folder.

Open Settings › Agents.
Create an agent, give it a name and instructions, and pick its tools and folder.
Save. The agent appears in the top bar.

If you skip this step, the default agent works in ~/Documents/PP Agent.

4. Start chatting

Pick an agent in the top bar, then type a message. Useful things to know:

Type / for slash commands and skills.
Drop files into the message box, or use the attach button.
Press Cmd+J (Ctrl+J on Windows and Linux) to open the side panel with a terminal, files, and a browser.
Read-only actions run freely. Saving files, shell commands, and actions that change data in your systems ask first. See Tools and approvals.

A good first task

Try something that only reads, so you can see the flow without approving changes:

Read the attached Q3 report and list the three biggest changes from Q2,
with the page each number comes from.

Then try a task that writes, such as creating a Word file or updating a spreadsheet, and watch the approval step.