Chat with documents
Drop in PDF, Word, Excel, PowerPoint, OpenDocument, RTF, EPUB, HTML, or CSV files and paste images. Mermaid diagrams and SVG previews are drawn in the chat; HTML pages and charts run in a sandbox.
Desktop AI agent · Open source
An AI agent for the whole business team.
Chat with your documents, put your business systems to work through MCP, and switch between agents built for each job. Nothing saves a file or changes a system without asking first.
macOSWindowsLinuxFree and open source · Apache-2.0
Runs on your computer
Keys in your OS keychain
Read-only work runs freely
Changes ask first
Why PP Agent
The power of a coding agent, without the terminal.
Coding agents are powerful but built for developers. PP Agent puts the same engine, the Pi SDK, behind an interface that people in finance, sales, operations, or HR can use: plain-language settings, no terminal, and nothing that writes a file or calls a business system without asking first.
A tech team sets up the models, tools, and connections once. Everyone else works through ready-made agents that already know the rules, the systems, and the folder for their job.
Read-only actions run freely. Saving files, shell commands, and anything that changes business data ask first — with a policy for every tool.
Features
One agent, the way your team already works.
Documents, business systems, research, and long-running work share one conversation, one workspace, and one approval gate.
Your business systems
Connect MCP servers with OAuth sign-in, or paste a configuration from Claude Desktop or Cursor. Each agent sees only the servers and tools you give it.
Purpose-built agents
Agents carry their own rules, tools, skills, model, starter prompts, and workspace folder. Switch mid-conversation, or hand one to a colleague as a .json file.
Deep research
/research frames the question, confirms a plan, researches in rounds with parallel helpers, rates each key claim, red-teams the draft, and saves cited Markdown and HTML reports.
Work that takes hours
/goal keeps the agent working turn after turn until the goal is done, with a checklist pinned above the message box and up to five read-only research helpers.
Approvals you control
Read-only actions run freely. Saving files, shell commands, and changes in your systems ask first, and every tool can be set to automatic, ask, allow, or off.
Any modelAnthropic, OpenAI, Google, OpenRouter, Mistral, Groq, or your own gateway.
Office filesBuild and edit Word, Excel, and PowerPoint files with the bundled OfficeCLI.
Web searchSearXNG, Serper, Tavily, Brave, Exa, webclaw, or DuckDuckGo, with fallback.
Side panelTerminal, workspace files, and a browser with separate sign-in profiles.
Desktop controlRead dashboards freely; clicking and typing in apps ask first, on macOS and Windows.
Scheduled tasksRun a prompt daily, weekly, or at an interval; each run is saved as a chat.
Memory and skillsNotes between chats, imported Agent Skills, and /skill commands.
Voice inputA local whisper model, or the speech-to-text service you choose.
Telegram and ZaloChat with the agent and answer approvals from the messaging app you use.
Remote accessUse the app from your phone or another computer on the same Wi-Fi.
History you ownSearch, pin, branch, and export chats to Markdown or HTML.
Private by designEncrypted keys, local conversations, and no account to create.
Get started
Set up once. Then just ask.
No server to run, no account to create. The app checks for updates and keeps working in the background when you close the window.
Connect a model
Sign in to Anthropic, OpenAI, Google, OpenRouter, Mistral, Groq, and more with an API key, or add your team's OpenAI- or Anthropic-compatible gateway.
Add your systems
Add MCP servers, create agents, install skills, and choose which tools each agent may use. Paste configurations you already use in Claude Desktop or Cursor.
Start working
Pick an agent, attach a document, and ask. Approve the changes you want, set a standing policy for the rest, and hand long work to /goal or/research.
Security and privacy
Designed for work that has consequences.
PP Agent is a local-first desktop app. Conversations and files stay on your computer, except what you send to the model providers, search services, and MCP servers you configure.
- Nothing changes silently. Writes, shell commands, and system actions pass an approval gate. Each tool can be automatic, always ask, always allow, or off.
- Keys stay in your OS keychain. Provider keys, search keys, and bot tokens are encrypted with a master key kept outside the app's data folder.
- You choose what leaves your computer. Search queries go to the service you pick, and pages on private networks are never opened.
- Guardrails stay on the computer. Settings that let the agent run programs without asking cannot be changed from a phone.
Install
Download PP Agent
Installers for every platform are on thelatest GitHub release, with SHA-256 checksums in SHA256SUMS.txt.
| System | File |
|---|---|
| macOS (Apple silicon) | pp-agent-<version>-mac-arm64.dmg |
| macOS (Intel) | pp-agent-<version>-mac-x64.dmg |
| Windows | pp-agent-<version>-win-x64.exe |
| Linux | pp-agent-<version>-linux-x86_64.AppImage or …-amd64.deb |
Updates and signing. Windows, Linux AppImage, and Developer ID-signed macOS builds update from inside the app. Unsigned builds can still show warnings: on macOS, run xattr -dr com.apple.quarantine "/Applications/PP Agent.app" once if the app is blocked; on Windows, choose More info → Run anyway in SmartScreen.
Give your team an agent that asks first.
Bring your documents, your business systems, and the way your team already works. PP Agent handles the rest on your own computer.