Plandex

Terminal AI coding agent built for large projects, with tree-sitter project maps and a reviewable diff sandbox.

Open SourceSelf Hosted
0.0 (0)

About

Large refactors and multi-step features are the target workload for Plandex, a Go-based terminal agent that maintains an effective context of 2 million tokens and indexes projects of 20 million tokens or more using tree-sitter maps covering over 30 languages. Instead of editing files directly, it accumulates proposed changes in a version-controlled sandbox where cumulative diffs can be reviewed, revised, or rejected before anything touches the working tree, and autonomy is configurable from full auto down to step-by-step approval. Model packs mix providers, so Anthropic, OpenAI, Google, and open weights models each handle the roles they fit best, with billing through your own API keys. The MIT-licensed server self-hosts with Docker; the hosted Plandex Cloud began winding down in October 2025 and development activity has slowed since, so self-hosting is the practical path today. The repository has about 15,500 GitHub stars, and it remains a reference design for diff-first, large-context agent workflows.

Should you use Plandex?

Pick it when

Pick it for large refactors across big codebases where you want every proposed change held in a reviewable sandbox until you approve it, and you are comfortable self-hosting the server with Docker and mixing models from several providers by role.

Look elsewhere when

Skip it if you need an actively developed tool or a hosted option: Plandex Cloud began winding down in October 2025 and the repo has had no pushes since, so you run the server yourself. Aider or OpenCode need less setup.

Alternatives to Plandex

  • Aider

    Commits each edit straight to git instead of staging it in a sandbox, installs with pip, and needs no server, at the cost of less review before changes land.

  • OpenCode

    Actively developed with one-line installs and no server to host, though it is not built around staging diffs at multi-million-token scale.

  • Serena

    Adds symbol-level navigation and safe renames to the agent you already use via MCP, a lighter way to handle big codebases than adopting a new agent.

Reviews (0)

Leave a Review

No reviews yet. Be the first to review!