# Decomposer - Full Documentation for LLMs

> Turn messy goals into clear, executable plans with honest AI guidance.

Website: https://decomposer.io
Built by: Skylark Creations (https://skylarkcreations.com)
Founder: David Kooi (https://davidkooi.com)

## Overview

Decomposer is a web application that transforms unstructured goals, project briefs, or messy ideas into structured, actionable task plans. It uses an AI decomposition pipeline to break down complex objectives into concrete steps, each labeled with who should do it: you (Human), AI tools (AI), or a professional (Expert).

The app is designed as "the thing you use before you use ChatGPT" -- it helps users figure out what to do, in what order, and with what tools, before they start executing.

## How It Works

### Input
Users provide:
- A goal or project description (free text)
- Optional: supporting documents, images, or files
- Optional: override assumptions and constraints

### Pipeline
The system runs a multi-stage AI pipeline:
1. Classifier -- Categorizes the goal type and complexity
2. Assumptions -- Extracts implicit assumptions and constraints (skipped if user provides overrides)
3. Decomposer -- Breaks the goal into 8-14 actionable task cards
4. Synthesis -- Creates a strategic overview connecting all tasks
5. Evaluator -- Scores plan quality across 5 dimensions

### Output
Each plan contains:
- Goal statement -- Clarified version of the user's input
- Assumptions and constraints -- What the plan assumes to be true
- Task cards (8-14 per plan), each with:
  - Title and description
  - Label: Human, AI, or Expert
  - Execution model: ai_autonomous, decide, create, configure, execute, or consult_expert
  - Dependencies (blocking, informational, artifact)
  - Input requirements
  - AI prompt (for AI-labeled tasks) -- copy-ready for ChatGPT, Claude, etc.
  - Risk notes
  - Status tracking (not_started, in_progress, done, blocked, skipped)
- Synthesis -- Strategic narrative connecting all tasks
- Quality score -- 5-dimension evaluation (coverage, redundancy, actionability, label routing, synthesis fit)

## Use Cases

Decomposer works for any complex goal that benefits from structured planning:

- Launch projects -- Start a business, plan an event, ship a product
- Learn skills -- Structured learning paths for coding, cooking, languages, etc.
- Research and analysis -- Break down research into phases with clear deliverables
- Creative projects -- Write a book, design a website, produce content
- Home and DIY -- Renovations, moves, garden planning
- Career transitions -- Plan job changes, skill development, portfolio building

## Pricing

- Browse Examples -- No account required. View sample plans and results.
- Free Account ($0/forever) -- All features included. Up to 3 plan generations.
- Pro ($10/month) -- All features included. Unlimited plan generations. Cancel anytime.

## Technical Details

- Built with Next.js 15, React 19, TypeScript, Tailwind CSS 4
- Uses frontier AI models for plan generation
- Plans are private by default
- Shared plans use unlisted links (not indexed by search engines)
- Data is not used for AI training
- Supports file uploads (documents, images) as planning context

## Key Differentiators

1. Structure, not soup -- Actionable cards instead of paragraphs
2. Honest labels -- Know what is yours, what AI can handle, what needs an expert
3. Copy-ready prompts -- Every AI task includes a prompt tuned for that step
4. Dependencies and risks -- See which steps block others and what could go wrong
5. Synthesis -- Never lose the forest for the trees
6. Persistent plans -- Plans that outlive a chat window, with sharing and export
7. Tool matching -- Pairs each task with relevant AI tools from the agentic.ai directory

## Integrations

### Agentic.ai (https://agentic.ai)

Decomposer integrates with the agentic.ai directory to recommend AI tools for each task in a generated plan. From any task in the focus view, users can request curated recommendations sourced from agentic.ai's directory of agentic AI products. Each recommendation includes the tool's tagline, agenticness score, deployment model, and pricing model, with a direct link to the full listing on agentic.ai.

The integration is intent-aware: a task to "find clients for a small business" surfaces sales-outreach tools; a task to "build a stock prediction module" surfaces coding agents. Users can rate recommendations as helpful or not helpful, providing feedback that improves the recommender over time.

This makes Decomposer not just a planner, but a planner + tool-chain orchestrator: it tells you what to do, in what order, AND which AI tools can do it for you.

## API

Decomposer does not currently offer a public API. All functionality is accessed through the web interface at https://decomposer.io.

## Contact

- Website: https://decomposer.io
- Privacy Policy: https://decomposer.io/privacy
- Terms of Service: https://decomposer.io/terms