Welcome to AITaskManager (AITM). This guide walks you through everything you need to know to get real work done with it, from your first task to reading the results.

Overview & Getting Started

What is AITM?

AITM is a desktop app that turns a single English prompt into working, tested, reviewed code. You describe what you want in a Task, AITM runs it through a 9-step pipeline — branch, architect, code, review, fix, test, e2e, docs, merge — and hands you back a finished change on its own git branch, ready to merge.

AITM vs. using Claude Code or Cursor directly

Tools like Claude Code or Cursor are excellent for one AI session at a time, driven interactively by you. AITM sits a level above them: it orchestrates those same AI coding assistants for you, automatically.

Every task runs in its own isolated git worktree, so parallel tasks never step on each other's files. Each task moves through the full pipeline without you babysitting the AI — the same AI session that writes the code also gets reviewed, fixed, and tested before anything reaches your main branch. A watchdog detects stuck or "zombie" AI processes and automatically retries them, and finished work is auto-merged into your dev branch once it passes review and tests.

In short: Claude Code and Cursor give you an AI pair programmer. AITM gives you an AI pair programmer that queues its own work, checks its own work, and only interrupts you when something needs a human decision.

Two Ways to Get a Project In

Entry point	Use it when
Import an existing project	You already have a working repo/folder on disk and want AITM to manage it.
New Project Wizard	You're starting from nothing — the wizard scaffolds a fresh project (config, CLAUDE.md, agent files, .gitignore) in 7 steps and hands you an importable project at the end.

Either path ends the same way: a project AITM knows about, added in Settings (or via the wizard on first run) with a Name, Path, Tab (a dashboard grouping), a default AI Provider, and a Default branch (usually dev).

Find Your Way Around the Dashboard

Once a project is ready, everything happens from the Dev Dashboard:

• Tabs across the top group related projects (e.g. all the pieces of one product).
• Project cards show Git state, quick actions, and a + Task button.
• The queue strip shows what's pending, running, or waiting for you.
• Chat and Brainstorm buttons sit next to every project for talking to the AI without touching code.

Core concepts

Concept	In one sentence
Project	A codebase AITM manages — imported once, then always available in the dashboard.
Task	A single unit of work you assign — "add a button", "fix a bug", "implement an endpoint".
Pipeline	The automatic multi-step process a task goes through: plan → code → review → fix → test → merge.
Agent	The AI "worker" running one pipeline step (e.g. the agent that writes code, the agent that reviews it).
Contract	A saved fact sheet describing how part of your project behaves, so the AI doesn't have to rediscover it every time.

Who AITM is for

AITM fits you especially well if you:

• Manage multiple projects at once, possibly in different languages/frameworks
• Want to use AI systematically and repeatably, not just as an occasional smart autocomplete
• Need visibility into what the AI is doing, with the ability to step in when needed
• Want the AI to remember your project's conventions instead of re-explaining them in every conversation

Your Getting-Started Journey

Here's the path most people take from a blank project to a merged change, with a link to the full detail for each step.

Step	What you do	Learn more
1. Import your project	Point AITM at your existing local git repository; onboarding maps the codebase automatically (see below).	—
2. Create your first task	Three ways in: the New Task modal (fastest, for a task you've already thought through), Chat, or Brainstorm. In day-to-day use, most tasks start in Chat or Brainstorm — you shape the idea conversationally, then it becomes a task.	Creating a Task · Chat · Brainstorm
3. Plan first, optionally	Not sure yet what you want? Discuss it in Chat or Brainstorm before committing to a task — neither ever edits your code directly.	Chat · Brainstorm
4. Watch the pipeline run	Architect plans, Code implements, Review and Fix loop until it's clean, then Test and E2E run in parallel before merge.	The Pipeline
5. Skills & Variables guide the AI	Reusable convention cards (Skills) and named references to parts of your app (Variables) keep every step consistent with your project.	Skills · Variable Explorer
6. Pipeline task vs. Agent task	A pipeline task runs the full 9-step reviewed/tested workflow; an agent task runs the same AI autonomously without the structured review/test gates, for quick self-directed work. Chat and Brainstorm let you pick either when converting a conversation into a task.	The Pipeline
7. Automated testing	Unit tests and end-to-end browser tests run automatically as part of the pipeline, in parallel, against an isolated instance of your app.	Testing
8. Running tasks in parallel	Multiple tasks can run at once, capped by how many concurrent pipeline "slots" your machine's CPU/memory can comfortably handle — configurable in Settings.	Settings
9. If something fails	The watchdog retries a stuck step automatically first. If it still fails, resume the task from the failed step rather than starting over — or open Chat to discuss the failure with the AI before deciding how to proceed.	Reading Results & Archive

Chat is for quick, contract-aware Q&A and small planning in one ongoing conversation; Brainstorm is for open-ended exploration before you commit to a task. Neither ever edits your code directly — both only ever produce a task, which then runs the real pipeline.

Running Fast Without E2E

E2E is the slowest pipeline step — it builds and boots the app, then drives it through a real browser. While iterating, keep the pipeline's reliability setting below the E2E threshold to skip it (and, at the lowest tier, unit tests too) for faster small tasks. Once a batch of related changes has landed, create one separate integration task with reliability turned up to run the full E2E regression against them.

Importing a project

Before AITM can manage a project, it needs to "get to know" it. Adding a project triggers an automatic, non-destructive, and free (no AI cost) static pre-scan: stack & language detection, a code map of file dependencies, UI element extraction for frontends, variable discovery, and documentation linking. You can safely re-run this pre-scan any time.

First run

When you open AITM for the first time, add a project (point it at a local git repository), then click New Task to describe your first change. AITM creates a branch, spins up an isolated worktree, and starts the pipeline. You can watch progress live on the dashboard, and you'll find the finished result waiting for you once the pipeline completes.