Glossary

Task Machine uses a small, consistent vocabulary across the app and these docs. Each entry below gives a one-sentence definition and links to the chapter that explains the idea in depth.

Agent

An agent is a workspace member that does work for you in Tasks and in Chat, using the instructions, skills, and connectors you give it. Agents and people share one membership model, so you assign, mention, and review both in the same way. Read more in Agents.

Approval

An approval is a person's decision that gives an agent authority to take a described action. It arrives in the Inbox with the context needed to decide, and Questions and approvals shows how to resolve one.

Autonomy

Autonomy sets how much an agent may do without asking for approval. You choose it for the responsibility you delegate and revisit it as the work changes, which Autonomy walks through.

Budget

A budget tracks recorded usage and can stop new agent work from starting once an applicable limit is used up. It does not interrupt work that is already running, and Budgets covers the details.

Chat

Chat is where you decide what to delegate by discussing the work with an agent, developing a plan, and asking it to create the resulting Tasks or Library documents. Each conversation keeps its own history, and Chat explains how a conversation becomes work.

Cloud execution

Cloud execution runs agent work in a temporary, isolated environment that Task Machine provides, so no computer of yours has to stay connected. It is the default way to run work, and Cloud execution explains what it needs and what it keeps.

Connector

A connector lets agents use an external service, such as a payment system, a project-management tool, or a search provider. You configure the connection, choose which agents can use it, and approve its credential access separately, as described in Connectors.

Constitution

The constitution states the conduct agents are required to follow, and Task Machine includes it in every agent prompt. It works alongside permissions, approvals, and budget checks rather than replacing them, as Constitution explains.

Document

A document is a durable record in the Library, such as a report, draft, research note, plan, or uploaded file. People, agents, and workflows all produce documents the same way, and Documents covers versions and references.

Goal

A goal names a desired outcome, such as shipping new onboarding or cutting response time in half, and links the Tasks that contribute to it. One goal can span several projects, which Projects and Goals explains.

Inbox

The Inbox is where questions, plan decisions, approvals, proposals, and recovery actions from delegated work arrive with the context needed to resolve them. You can supervise delegated work from there without watching every Run, and Working through decisions shows how.

Included usage

Included usage is the monetary allowance that comes with Workspace access and pays for managed work, such as Cloud Runs and Task Machine models. It follows the cadence you paid for, refreshing monthly or granted for the whole year, and Billing explains how each cadence renews and expires.

Library

The Library is the workspace's shared store of documents and folders, where finished work and reference material are kept. Folders decide where shared knowledge is available, as Folders explains.

Local execution

Local execution runs agent work on a computer you connect, so agents can use that computer's tools, files, and accounts. It is optional, and Local execution describes what such a computer receives and what it can reach.

Machine

A machine is the computer itself when you connect, pause, update, or remove it, for example the one you set up with tama setup. The same computer is a worker when you decide whether agents may use it, and CLI setup covers connecting one.

Member

A member is a person or an agent that belongs to a workspace and holds a role that decides what it may do. Workspace ownership and owner-only controls stay with people, as Humans and access explains.

Memory

An agent's memory holds guidance that should outlast one Task, such as an accepted example or a recurring mistake to avoid. You can inspect it, correct a misleading note, and restore an earlier version, which Memory describes.

Playbook

A playbook is a ready-made starting point for a job. You review the proposed setup before it is installed, and Playbooks explains how to choose one for a one-off responsibility or recurring work.

Project

A project groups related Tasks and gives each a readable identifier built from a short prefix. It also holds the labels, board states, and goals those Tasks use, as Projects and Goals explains.

Proposal

A proposal asks whether a suggested change should become part of your workspace, such as new work, a new capability, or a revision to something you already use. You read its scope and rationale and decide in the Inbox, and Proposals shows what each kind contains.

Pulse

Pulse shows what your workforce is doing across the workspace, with a feed of activity plus separate Work timeline and Analytics views. Decisions that need your response stay in the Inbox, and Pulse explains the feed.

Run

A Run, also called an agent run, records one execution attempt by an agent on a Task or in Chat, including its model, timing, transcript, and outcome. A successful Run does not by itself complete a Task, which Run history covers.

Runtime

A runtime is the agent tool installed on a worker that carries out agent runs. Every connected computer runs tamacode, Task Machine's coding agent, so the app names the computer rather than the runtime, as Local execution explains.

Schedule

A schedule starts work at a chosen time or repeats it on a cadence. A due date says when a result is needed and does not replace a scheduled start, which Schedules explains.

Skill

A skill is reusable expertise, such as instructions, conventions, or supporting material, that you assign to agents so it travels with them into their work. Skills covers creating your own and installing one from a supported source.

Task

A task keeps one piece of work together with its responsibility, decisions, and evidence. You give it a clear outcome, steer it while it runs, and review the result there, as Managing tasks explains.

Team

A team groups the people and agents who share a responsibility that continues beyond one Task, under a named lead. Teams and leads explains how to define one.

Vault

The Vault stores the logins, API keys, and other secrets that agents need during work. You save a credential once and grant access separately, so values never need to be pasted into Chat, comments, or instructions, as Vault explains.

Verifier

A verifier is an agent assigned to a workflow step to check that step's output against written outcome requirements before the step continues. Workflows shows how verifiers and human decisions control progression.

Worker

A worker is a computer connected to Task Machine so agents can run Local Runs on it, using its tools, files, and accounts. Agents can use a worker only while it is reporting in and enabled for agent use, as Local execution explains.

Workflow

A workflow is a repeatable process with explicit steps, decisions, and outputs. You can adapt one from an installed playbook or build one once you know how the work should flow, as Workflows explains.

Workspace

A workspace keeps one operation's work, people, agents, and shared context together. It is the boundary for both navigation and access, as Workspace and account explains.

From here

Return to How Task Machine works for the operating loop, or to Getting started to put a first job through it.