Project Starters
Start new projects from pre-configured starters that include rules, dev server settings, and starter files.
Why Project Starters matter
Setting up a project from scratch means configuring rules, dev server commands, and starter files every time. A Project Starter captures that setup once so you can reuse it. They are especially valuable when you have a proven project structure you want to repeat, like a Next.js app with your preferred linting rules and folder conventions, or a Python data project with your standard notebook layout.
What a Project Starter includes
A starter can bundle any combination of:
- Starter files - the initial codebase or folder structure that gives you a working starting point.
- Rules - pre-configured instructions so Panda already knows the conventions for this type of project.
- Dev server settings - the correct command and port for the framework, so the live preview works immediately.
- Recommended skills - Skills that are particularly useful for the project type, such as database skills for a full-stack app.
Choosing a Project Starter
When creating a new project, the first wizard step offers Start Fresh at the top, an empty folder with no generated files, followed by the built-in Project Starters, grouped by category.
| Starter | Category | Use it for |
|---|---|---|
| Personal Assistant | Productivity | Running your working day out of a single status.md, a sweep across mail, calendar, tasks and chat, and a capture pass that records what was decided |
| Knowledge Map | Productivity | Thoughts, notes, decisions and research trails, kept as linked Markdown |
| Full-Stack App | Development | A web application with a database and a backend |
| Website | Development | Static sites with no database and no backend, such as a landing page, a personal homepage, a docs site, or a quick local tool |
| API / Backend Service | Development | FastAPI, Express, NestJS and friends, with an OpenAPI, GraphQL, gRPC or tRPC contract |
| Desktop Electron App | Development | A cross-platform desktop app, packaged and optionally auto-updating |
| MCP Server | Tools | Tools, resources and prompts exposed to Claude Desktop, Claude Code, PandaOS or another client |
| Analyze Data | Data | Notebooks and queries against a database, Supabase, or local files |
| Data Warehouse | Data | dbt and SQL pipelines with a layered warehouse architecture |
PandaOS App appears under Tools with a Soon badge and cannot be selected yet.
Pick one and the wizard adapts to it. Starters with options get a Configure step; those that link to Supabase or a database get a step for that too; the Personal Assistant and Knowledge Map each add an optional Connect step, for the apps the assistant should watch and for where your notes and documents already live; Website skips all of them, since a static site has neither a database nor connectors; Start Fresh skips straight to choosing agents. Once created, the project opens fully configured with the framework detected, the dev server ready, and starter rules applied. You can always customize everything after creation.
The Personal Assistant starter
The Personal Assistant starter is built for running a working day rather than building software. Everything in it revolves around one file, status.md at the project root, which holds what needs doing today, what you are waiting on, your open loops, the decisions you have made, and what was recently finished.
Two skills keep that file true, from opposite directions:
- orchestrate runs at the start of the day. It sweeps whatever you have connected — mail, calendar, tasks, chat, boards — works out what actually needs you, and rewrites the top of the file. It also tells you what it could not reach, so a brief that is missing half your day never looks like a quiet one.
- capture runs when you wrap up. It reads back what happened, separates what was decided from what is still open, asks about anything ambiguous instead of guessing, and records the outcome. In a team project it also routes other people's action items to your board instead of burying them in a personal file.
A third skill, brief, loads the current state at the start of a session so you are not re-explaining yourself. All three come with the Chief of Staff agent, which the starter hires for you. You can also hire it into any other project from the agent browser, or install the skills on their own from the skills catalog.
Setup choices
Rhythm decides the shape of the loop: a Daily Brief (a morning sweep and an end-of-day write-back), Continuous (capture at the end of every session, no fixed ritual), or a Weekly Review (light daily passes with one real reconciliation per week).
Who's involved decides where captured work goes. Just me keeps every action item in your own file. Me and my team adds the routing rules, sweeps team surfaces like mentions and your board columns, and sends team action items to the board with a named owner rather than into a file nobody else reads.
Connecting apps
The Connect step groups apps by the job they do — email, calendar, tasks, notes, files, and for team projects chat and boards — rather than by vendor. Connect one per row and the assistant uses whatever is available when it runs. The whole step is optional: skip it and connect the same apps later from Apps, and nothing about the project changes.
What stays yours
The skills are deliberately generic. Anything personal — who you are, your working hours, which chat spaces are worth reading, whose board columns are whose — lives in knowledge files under .pandaos/knowledge/, which the assistant fills in by interviewing you on the first run. That split is what lets the same starter work whether you are on Gmail and Slack or Outlook and Teams.
The curated list of what to watch matters more than it sounds. Without one, a morning brief includes every dead channel and unread newsletter, and stops being read within a week.