Plandex.
Terminal-based AI coding agent with a plan-first methodology.
What it is.
Danny Holloway's open-source terminal agent. Plandex's distinguishing move is the plan stage — the agent drafts a structured plan, the developer reviews and edits it, and only then does the agent execute. Each plan runs in a sandboxed branch; changes apply only after explicit approval. Self-hostable end to end.
Where it fits.
Engineers running large refactors or migrations where uncontrolled agent execution is the risk. The plan stage forces the agent to expose its reasoning before touching files — useful when the codebase is too big to review every diff line by line. Self-hosted deployment appeals to teams that cannot send proprietary code to managed services.
- Plan stage forces agent reasoning to surface before execution
- Sandboxed branches prevent destructive changes by default
- Self-hostable end to end with no external dependencies
- Terminal-first UX excludes non-engineer reviewers
- Plan stage adds latency that mass-market tools skip
Frequently asked.
What does plan-first mean?
Plandex drafts a structured plan, presents it for review, and only executes after the developer approves. Closer to an architect review than a free-running agent.
Can Plandex run self-hosted?
Yes. The full stack is open source and runs in Docker. Buyers point it at any OpenAI-compatible model endpoint, including local Llama and Mistral servers.
How does Plandex compare to Aider?
Both are terminal-native and open source. Aider commits each edit to git atomically; Plandex stages every edit in a sandboxed branch and waits for explicit approval to merge.
