Prospero

Key facts: Prospero, announced on October 31, 2026 by Chris Wellons (nullprogram), is a native coding harness written in C++ with a Dear ImGui interface. Its claim is striking: around two orders of magnitude fewer resources than typical web-stack harnesses, without giving up subagents, Git worktrees, MCP support or desktop control for testing GUI apps. Configuration lives in a plain INI file, and it ships with no model providers built in: you connect llama.cpp, OpenRouter, Anthropic or OpenAI APIs yourself. You cannot use a Claude subscription with it.

Plans: Free Community Edition on GitHub under the Unlicense, with a reminder nag and no built-in headless engines. The paid Professional Edition, sold on itch.io for a nominal fee, is DRM-free with free updates and adds headless engines for remote deployment.

Best for: Developers who want a fast, light harness and do not mind bringing their own models. Who should skip it: anyone who wants a subscription model included, or a zero-config hosted product.

Pros

• Around 100x fewer resources than typical web-stack harnesses, in a native C++ app
• Subagents, Git worktrees, MCP and desktop control for GUI testing included
• Plain INI configuration instead of JSON, easy to check into source control
• Free Community Edition, with a tiny 5MB-class binary

Watch out for

• No model providers built in: you supply your own API or local model, and Claude subscriptions do not work
• Community Edition lacks headless engines and shows a reminder nag
• A one-author project, so support and roadmap depend on a single maintainer

Best for

Developers who want a fast, light harness and do not mind bringing their own models. Who should skip it: anyone who wants a subscription model included, or a zero-config hosted product.

Pricing

Free Community Edition on GitHub (Unlicense). Professional Edition sold on itch.io for a nominal fee, DRM-free with free updates.

FAQs

What is a coding harness? The layer that lets an AI agent read files, run commands and use tools in a loop, with a chat interface on top.

How light is Prospero? Its author claims around two orders of magnitude fewer resources than typical web-stack harnesses.

Which models does it use? None built in: connect llama.cpp, OpenRouter, Anthropic or OpenAI APIs yourself. Claude subscriptions do not work.

What is the difference between editions? Community is free with a nag and no headless engines; Professional adds headless engines for remote deployment.