ByHeather MacAvelia·Last verified Jun 20, 2026

Plandex

Name: Plandex
Author: Plandex

★3.4/ 5

by Plandex

Independently Reviewed

Visit github.com →

Open-source terminal AI coding agent for large projects with 2M token context. Cloud service winding down. Self-hosted only. 15.5K GitHub stars. Free BYOK.

From

Custom

free ↗

GitHub

⭐ 15.5k

View on GitHub ↗

—

Rating

MCP

Compatible

Plandex is the open-source terminal-based AI coding agent designed specifically for handling large projects, large files, and large tasks. Released as open source, Plandex features a diff review sandbox, configurable autonomy modes (from full auto to granular step-by-step), and smart context management supporting up to 2M tokens of project context with directory indexing via tree-sitter project maps for codebases with 20M+ tokens. The project has 15.5k GitHub stars and 23 contributors. Note on current state: Plandex Cloud (the hosted commercial offering) began winding down on October 3, 2025 per the official README. The plandex.ai website returns a 404. Plandex continues as a self-hosted open-source project installed directly from GitHub. Teams evaluating Plandex should be aware they are adopting community-maintained tooling without commercial backing rather than a productized service. Pricing follows the BYOK open-source model: Plandex itself is free, with users paying only for the LLM API usage they configure. Plandex supports Claude, OpenAI GPT, and Google models, with context caching used across the board for OpenAI, Anthropic, and Google models to reduce costs and latency. Curated model packs offer different tradeoffs of capability, cost, and speed, including open source and provider-specific packs. Plandex differentiates from Aider, Claude Code, and Cursor through explicit focus on large-task handling: rather than optimizing for quick single-file edits, Plandex plans and executes multi-hour development tasks across many files, with diff review checkpoints that let developers approve incremental changes rather than reviewing massive end-state diffs. The platform supports 30+ programming languages, project-aware chat mode for fleshing out ideas before implementation, and automated debugging of terminal commands including builds, linters, tests, deployments, and browser applications via Chrome. Integration depth is limited to LLM providers and Git. Plandex connects to OpenAI, Anthropic, and Google as model providers. It operates as a terminal tool within existing Git repositories. There is no IDE plugin (VS Code integration is in development but not shipped), no MCP support, no connections to project management tools, Slack, or other collaboration platforms. Teams needing rich IDE integration or app ecosystem connections will find significant gaps compared to Cursor or OpenAI Codex. With the cloud service winding down, teams needing a stable hosted autonomous coding tool should evaluate Claude Code ($20/month) or Cursor ($20/month) instead. Plandex remains a viable self-hosted option for power users comfortable with terminal workflows who want fine-grained control over large multi-file tasks. Current state Q2 2026: Plandex is at version 2.2.1. The cloud wind-down began October 2025 and the commercial website is offline. The project remains active on GitHub with recent commits. No security certifications exist. MCP is not supported. The project is community-maintained open source.

Pricing

free

View pricing ↗

Segment

b2b

Setup

moderate

Verified

Jun 20, 2026

Transparency

Public

Contract

Month-to-month

Data training

Not Trained

Autonomy

Human Optional

Capabilities

agentic-codingterminal-agentmulti-file-editinggit-nativeopen-source

Pros & Limitations

Editorial assessment

Pros

✓Designed specifically for large tasks and codebases: 2M token context management with tree-sitter project maps for 20M+ token codebases, multi-file diff review checkpoints, and configurable autonomy are purpose-built for hours-long development tasks that other AI coding tools struggle to handle.
✓BYOK with context caching across OpenAI, Anthropic, and Google: pay only for actual LLM API usage with cost optimization built in, curated model packs for different capability and cost tradeoffs, and full self-hosting under open-source license with no vendor lock-in.
✓Configurable autonomy from step-by-step to full auto with automated debugging: developers choose oversight level based on task complexity, and Plandex can automatically debug terminal commands including builds, linters, tests, and browser applications via Chrome.

Limitations

⚠Plandex Cloud is winding down and the main website is offline: users must install and run Plandex from the GitHub repository directly with no hosted service available, which is a meaningful operational step up from commercial alternatives.
⚠Terminal-only interface with no IDE plugin: Plandex requires command-line comfort and lacks visual IDE integration; VS Code support is in development but not shipped, limiting accessibility compared to Cursor or GitHub Copilot.
⚠Community-maintained project without commercial backing: the cloud wind-down reduces investment in the project, meaning fewer pre-built workflows, slower feature development, and higher adoption risk compared to funded alternatives like Claude Code or Devin.

Technical Details

Deployment

cliself-hosted

Model architectureOpen Source

Avg setup timeUnder 30 minutes: install via npm or homebrew, configure LLM API key, initialize first plan in repository

Autonomous rateConfigurable: Plandex supports full auto mode for autonomous task execution and granular step-by-step mode for human-in-the-loop oversight, with automated debugging of terminal commands

Integrations

OpenAIAnthropicGoogleGit