Concepts
Retask is a task manager with AI Agents built in. The idea in one line: you do what matters, and AI Agents do the rest.
In most setups, your to-do list lives in one tool (Jira, Linear, a notebook) and the AI that actually does the work lives somewhere else (Claude Code, Codex, Cursor). The two never really talk, so every time an agent runs you re-explain the work and rebuild its environment from scratch. Retask puts both in one place: the task is the record of what needs doing, and an AI Agent can pick it up and do it inside a safe, reproducible workspace.
This page explains the handful of words you’ll see everywhere in Retask. You don’t need to be technical to follow it — read it top to bottom once and the rest of the guides will make sense.
How the pieces fit together
Section titled “How the pieces fit together”Think of it as a set of nested containers, plus the “engine” that does the work:
- A Workspace is your whole account. It holds everything below.
- Workspace Members are the people (and agents) who can see and work in it.
- Projects group related work — usually one project per app, product, or area.
- Tasks are the individual pieces of work inside a project.
- To get work done automatically, Retask adds three more pieces:
- An Agent is an AI worker you can assign a task to.
- A Sandbox is a live, isolated environment where the agent runs.
- A Session is one individual run of an agent inside a sandbox.
- A Sandbox Template is a reusable recipe for building sandboxes.
The short version: a Task says what to do. An Agent does it. The Sandbox is where it works. A Session is one attempt.
The concepts, one by one
Section titled “The concepts, one by one”Workspace
Section titled “Workspace”Your top-level home in Retask. A workspace holds your projects, your people, your billing, and your settings. Most teams have exactly one. If you consult for several clients, you might keep a separate workspace per client so nothing mixes.
In the app: the workspace name and switcher sit at the top of the screen.
Workspace Member
Section titled “Workspace Member”Anyone you invite into a workspace. Each member has a role that controls what they can do:
- Owner — full control, including billing and deleting the workspace.
- Admin — manage members, projects, and settings.
- Editor — create and change projects and tasks.
- Viewer — look but don’t touch (read-only).
Agents also show up alongside people in the members area, under their own tab — they’re workers in the workspace, just not human ones.
Project
Section titled “Project”A container for related tasks — typically one per app, service, or area of work (for example, “Retask Web App” or “Mobile App”). Each project has its own board, its own set of statuses, and its own members. You can keep a project private to specific people or open it to the whole workspace.
In the app: a project opens to a Kanban board — columns like Todo, In Progress, In Preview, and Done, with task cards you drag between them.
The single most important concept. A task is one piece of work — a feature, a
bug, a chore. It has a title, a description, a status, a priority, assignees, an
optional due date, and a short key like RWA-3 that everyone can refer to.
What makes a Retask task special: it’s not just a sticky note. A task can be handed to an AI Agent, which goes off and does the work and reports back to that same task. The task is the durable record — the “contract” — that survives long after any single agent run ends.
An AI worker you can assign tasks to, the same way you’d assign a teammate. You set one up by filling in a short form: a name, a role (for example, a task processor), when it runs, which integrations it can use, and how its environment starts. Everything the agent needs lives on the agent itself — there’s no separate template to attach. Behind the scenes it’s powered by a coding AI such as Claude Code or Codex; you configure it once and then just hand it work.
Sandbox Template
Section titled “Sandbox Template”The reusable recipe for an agent’s work environment. A template says: which code repositories to pull in, which environment variables and secrets to set, and which tools to start. For example, a template named “Claude Code · Retask web” might clone two repositories, install dependencies, and launch the agent ready to go.
Set up a template once, and every sandbox built from it starts out identical — so agents get a clean, predictable environment every single time.
Sandbox
Section titled “Sandbox”A live, isolated environment where agents actually run, built from a template. It’s like giving the agent its own private computer that’s walled off from everything else, so your code and credentials are never shared between workspaces. Retask offers two kinds:
- Cloud sandbox — created and managed for you by Retask.
- Private VM — your own machine, connected to Retask with the CLI, running the AI tools you’ve already signed in there.
A sandbox can host several sessions and can be running, idle, or stopped.
Session
Section titled “Session”One individual run of an agent inside a sandbox — a single attempt at some work. A session has a live terminal you can watch in real time, so you can see exactly what the agent is doing and step in if needed. When it’s finished, the session sticks around as a record you can review later.
A session can be seeded with a starting point — either a prompt (“fix the login bug”) or a link to a task — so the agent knows what it’s working on the moment it starts.