> ## Documentation Index
> Fetch the complete documentation index at: https://graph.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Glossary

> The terms graph's docs use, each with a short definition and a link.

## Agent

A named model with a system prompt, a [model role](#model-role) and a list of [tools](#tool) it may call, defined in an agent file. Agents run in `graph ask`, `graph chat` and the [workbench](#workbench), and [plans](#plan) call them as `agent__<name>` tools. See [Agents](/agents/overview).

## Agent network

Any composition of [agents](#agent), [plans](#plan) and [tools](#tool) that call one another. See [Core concepts](/getting-started/concepts).

## Agent step

The `agent` [control step](#control-step), which hands one [plan](#plan) [step](#step) to a model that calls [tools](#tool) until it can return a result. An agent step is defined inline in the plan. See [agent](/plans/reference/steps/agent).

## Ask step

The `ask` [control step](#control-step), which asks the person running the [plan](#plan) a question and uses the answer. See [ask](/plans/reference/steps/ask).

## Body

The [steps](#step) a `route` branch, `map` or `reduce` runs: a single call or an inline list of steps. See [Conditions and bodies](/plans/reference/conditions-and-bodies).

## Built-in

An [agent](#agent), [plan](#plan) or [tool](#tool) that ships with graph. See [Customize a built-in agent](/agents/guides/customize-built-ins).

## Catalog

Every [tool](#tool) available to an [agent](#agent) or [plan](#plan), each named with a prefix that says where it comes from, such as `user__` or `plan__`. See [Core concepts](/getting-started/concepts).

## Control step

A [step](#step) that steers how a [plan](#plan) runs instead of calling a [tool](#tool). It branches (`route`), narrows a list (`filter`), repeats over a list (`map`, `reduce`) or stops the plan (`exit`), or hands the step to a model ([`agent`](#agent-step)) or a person ([`ask`](#ask-step)). See [Control steps](/plans/reference/control-steps).

## Decision model

A model that answers a fixed question with a likelihood or a choice among options, instead of writing text. [Gates](#gate) using `decide` and `decision` [tools](#tool-kind) use one through the `decider` [model role](#model-role). See [Models and providers](/getting-started/models).

## File version

The schema version a config, [plan](#plan), [tool](#tool) or [agent](#agent) file carries in its `version` key, separate from graph's release version. Each graph release reads a range of file versions and upgrades older files as they load. See [File versions](/reference/file-versions).

## Finish

How a [plan](#plan) ends and what it returns: `silent`, an `output` map built from [step](#step) results, or a `solver` that writes an answer. See [`finish`](/plans/reference/plan-file#param-finish).

## Gate

The condition on an `exit`, `route` or `filter` [step](#step): a comparison of values, an `infer` yes-or-no verdict from a model, or a `decide` call to a [decision model](#decision-model). See [Conditions and bodies](/plans/reference/conditions-and-bodies).

## Handoff

One [agent](#agent) passing the conversation to another through a `transfer_to_<name>` [tool](#tool). The other agent answers from then on. See [Build a multi-agent network](/agents/guides/multi-agent-networks).

## Layer

One of the three places graph loads [agents](#agent), [plans](#plan), [tools](#tool) and config from. [Built-ins](#built-in) load first, then your global files in `~/.config/graph/`, then project files in `./.graph/`. A file in a later layer replaces one with the same name in an earlier layer. See [Core concepts](/getting-started/concepts).

## Model role

A name such as `chat`, `judge` or a role you define, which [agents](#agent), [steps](#step) and [tools](#tool) use instead of naming a model. `[models]` in config.toml maps each role to a model. See [`models`](/reference/configuration#param-models).

## Pack

A set of [tools](#tool), [plans](#plan) and [agents](#agent) installed from a git repository with `graph pack add`, such as `github` or `slack`. Each pack is a `[packs.<name>]` entry in config, and its tools are called as `<name>__tool__<tool>`. See [Packs](/packs/overview).

## Plan

A file of [steps](#step) wired together by their results. Each step's input draws on earlier results through [templates](#template), and [control steps](#control-step) branch, repeat and stop the flow. See [Plans](/plans/overview).

## Shape

The observed output schema of a [tool](#tool), with an example, recorded each time the tool runs. See [Improve what graph knows about a tool](/tools/guides/tool-shapes).

## Step

One entry in a [plan](#plan): a [tool](#tool) call, a [control step](#control-step), or a call to an [agent](#agent) or another plan. See [`steps`](/plans/reference/plan-file#param-steps).

## Subagent

An [agent](#agent) another agent calls as a [tool](#tool). The subagent runs, returns a result, and the caller keeps the conversation. See [Build a multi-agent network](/agents/guides/multi-agent-networks).

## Template

A `{{...}}` reference in a [step](#step), [tool](#tool) or prompt that splices in an input or an earlier result. See [Template language](/reference/template-language).

## Thread

The saved history of a conversation with an [agent](#agent) in `graph ask` or `graph chat`. Continue one with `--thread`. See [Manage threads](/agents/guides/threads).

## Tool

Anything an [agent](#agent) or [step](#step) can call: MCP server tools, your tool files, built-in and [pack](#pack) tools, and [plans](#plan) and agents exposed as tools. See [Tools](/tools/overview).

## Tool kind

What a [tool](#tool) file runs: `exec` runs a command, `prompt` calls a model, `reshape` rebuilds JSON without a model, and `decision` asks a [decision model](#decision-model). See [Tool file](/tools/reference/tool-file).

## Workbench

The full-screen terminal app opened by `graph workbench`, where you draft, run and debug [plans](#plan), [agents](#agent) and [tools](#tool). See [Workbench](/getting-started/workbench).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.