Overview

Pumkin is a local-first AI agent IDE for developers. You build, run, and audit AI agents on your own machine — they talk to models running locally through Ollama and use tools you connect through MCP. Nothing leaves your computer. No cloud, no API keys, no per-token billing.

This guide gets you from a fresh install to your first working agent. If you hit something it doesn’t cover, email hi@pumkin.app — one founder, real answers.

How it works

Pumkin is a desktop app. Under the hood it runs a local server that orchestrates three things:

  • Your models — served by Ollama, running entirely on your hardware. Pumkin never sends your prompts to a third party.
  • Your tools — connected through MCP (Model Context Protocol). An agent can read time, call a script, hit a local API, or use any MCP server you point it at.
  • Your history — every run, message, and tool call is stored in a local database on your machine, so you can audit exactly what an agent did and why.

An agent is a saved configuration: a model, a system prompt, and a set of tools. You give it a task, it reasons, calls tools when it needs them, and streams back its work as a run you can inspect step by step.

What you need

  • Windows 10 or 11 (64-bit). macOS and Linux builds are coming; your license covers them when they ship.
  • Ollama installed and running. This is what actually runs the AI models. Setup is covered on the next page.
  • At least 8 GB of RAM. This is a real floor, not a suggestion — see Install Ollama & models for which models fit which machines.
  • A few GB of free disk for the app and whatever models you pull.

The five-minute path

  1. Install Ollama and pull a model — start with llama3.2:3b.
  2. Install Pumkin and launch it.
  3. Create your first agent, pick your model, and run a task.
  4. Add MCP tools when you want the agent to actually do things.

New to local models? That’s fine. You don’t need to understand how models work to use Pumkin — you need Ollama running and one model pulled. The next page walks through both.

Start with Install Ollama & models →