What it does
Plandex is a terminal-based AI coding agent aimed at big projects. It can plan and carry out tasks that run over many steps and touch dozens of files, handles up to 2M tokens of context directly, and uses tree-sitter project maps to index directories of 20M tokens or more. Generated changes sit in a cumulative diff review sandbox, apart from your project files, until you apply them. Plans have their own version control with branches, and Git integration can write commit messages. Autonomy is adjustable from full auto, where it loads files, implements changes, runs commands and debugs, to step-by-step review. You can combine models from Anthropic, OpenAI, Google and open-source providers using curated model packs. It runs from a REPL or a scriptable CLI. The hosted Plandex Cloud was winding down as of 3 October 2025 and no longer accepts new users, so the route described now is self-hosting in local mode, with Docker or your own server, using an OpenRouter key or other provider accounts. Windows is supported only through WSL.
Capabilities
- Up to 2M tokens of direct context, with tree-sitter maps for larger directories
- Diff review sandbox that keeps changes separate until applied
- Branches and version history for every plan
- Configurable autonomy from full auto to step-by-step review
- Models from Anthropic, OpenAI, Google and open-source providers
- Automated debugging of build, test and script commands
When to use it
- Working on a large codebase where changes span many files
- Reviewing AI-generated changes before they reach your files
- Comparing different models on the same coding task
- Self-hosting an open-source terminal coding agent with your own provider keys
