Plandex

Plandex is an open-source terminal-based AI coding agent built specifically for large, multi-file, real-world software development tasks — the kind that other AI coding tools struggle with. Created and released under the MIT license by a small team led by founder Dane Schneider, Plandex focuses on the use case where a task spans dozens of files, requires multiple sequential steps, and demands careful human oversight before changes are applied. The key insight: instead of applying AI changes immediately, Plandex stages all generated changes in a cumulative diff sandbox for review, so developers inspect the full set of proposed changes across every file before committing anything to the filesystem.

Plandex introduces plans as first-class objects — every task is a plan with its own context, branches, conversation history, and pending changes. Plans support git-like branching to try alternative approaches, version control with checkpoints for rollback, and incremental merging of changes. A 2M token context window handles large individual files, while tree-sitter project maps allow the agent to navigate and reason about codebases with 20M+ tokens across 30+ programming languages. Autonomy is fully configurable: full-auto mode runs continuously without confirmation, while step-by-step mode prompts for approval at each stage. Multiple developers can share a self-hosted team server with shared plan history.

Plandex Cloud (the original hosted service) began winding down on October 3, 2025 and no longer accepts new users. The maintainers concluded that self-hosting via Docker gives developers better privacy, control, and economics. The plandex.ai website no longer serves content, but the GitHub repository (github.com/plandex-ai/plandex) remains active with 15,400+ stars and the self-hosted path fully functional. Teams evaluating Plandex should understand they are adopting community-maintained tooling, which suits power users who want fine-grained terminal control over large multi-file tasks and prefer self-hosted infrastructure.

Plandex is completely free and MIT licensed. The CLI installs in one line with no dependencies. The server runs locally via Docker or can be deployed on any cloud provider (Railway, Render, your own VPS).

For model inference, you bring your own API keys from any supported provider (Anthropic, OpenAI, Google, OpenRouter) and pay provider rates directly with no markup from Plandex. OpenRouter is recommended as it gives access to all major providers from a single key.

Plandex Cloud (the original hosted service with centralized billing, multi-device usage, and no Docker setup required) began winding down on October 3, 2025 and no longer accepts new users. The self-hosted path is now the primary deployment option. For teams, running a shared Plandex server on Railway or a similar platform provides the equivalent of the original cloud experience with full data control.