What Plandex Can Do

Every feature built for
production code.

A deep look at the tools that make Plandex different from every other AI coding assistant.

Context Engine
2M tokens of real project understanding
Plandex doesn't just read the files you paste in. It indexes your entire repository using tree-sitter project maps, understanding functions, imports, and relationships — all within a 2M effective token window.
tree-sitter 2M tokens 20M+ dir indexing smart context mgmt
Effective context window2,000,000 tokens
Directory indexing20M+ tokens
Project map typetree-sitter AST
Avg files per task8–24 files
Context caching✓ enabled
Diff Sandbox
Review every change before it lands
All AI-generated changes are staged in a sandbox, not applied directly to your files. Review diffs across every modified file. Approve all, approve selectively, or roll back to any prior checkpoint — zero risk of silent overwriting.
staged diffs per-file review instant rollback full history
src/auth/jwt.ts — 3 changes
12 import { sign, verify } from 'jsonwebtoken'
13- const TOKEN_EXPIRY = '1h'
13+ const ACCESS_EXPIRY = '15m'
14+ const REFRESH_EXPIRY = '7d'
15
16- export function createToken(userId: string) {
16+ export function createAccessToken(userId: string) {
17 return sign({ userId }, SECRET, {
18- expiresIn: TOKEN_EXPIRY
18+ expiresIn: ACCESS_EXPIRY
19 })
20 }
Autonomy Control
Full auto to step-by-step — you decide
Enable full auto mode and let Plandex plan, code, run commands, test, and debug end-to-end. Or dial back to granular confirmation at each step. The level of autonomy is always in your hands, adjustable mid-task.
full auto mode step-by-step mode auto debug loops cmd execution
❯ plandex set-auto full
# Full auto: plan → code → test → debug
✓ Autonomy set to: full
 
❯ plandex "add Redis caching layer"
⟳ Planning... (8 steps)
⟳ Writing code... (6 files)
⟳ Running: npm test
✗ 2 tests failed — debugging
⟳ Fixing: cache key collision
⟳ Running: npm test
✓ All 47 tests passing
Model Flexibility
Mix the best models for each stage
Use Claude for reasoning, GPT-4o for speed, Gemini for long context, and open source models for offline work. Plandex routes each stage of the pipeline to the best-fit model — no single-vendor dependency.
Claude 3.5/4 GPT-4o / o3 Gemini 2.5 OpenRouter Ollama
claude-opus-4
Planning & architecture
PLAN
gpt-4o
Code generation
EDIT
gemini-2.5-pro
Large context review
EDIT
claude-sonnet-4
Test debugging
DEBUG

See it in your codebase.

Get started with a one-time payment and own Plandex forever. No subscriptions.

View Pricing →