Skip to content

Glossary

The terms this manual uses, in alphabetical order, with a pointer to where each is explained.

@mention
Writing @name in a chat message to send it to a particular crew member, or @crew to send it to all of them. See Crews.
Chat session
An interactive conversation started with --chat. The model remembers earlier turns until you /clear them. See Chat Sessions.
Checkpoint
A named point in a chat conversation, saved with /checkpoint, that you can go back to with /restore. See Chat Sessions.
Chief
The lead robot of a crew — the one that answers when you don't address anyone in particular. See Crews.
Config file
The YAML file holding your default settings, ~/.config/aia/aia.yml. A project can add ./config/aia.yml, and -c FILE substitutes another. See The Config File.
Consensus
With several models, an extra step that merges their answers into one response (--consensus). See Multi-Model & Consensus.
Context file
A file named on the command line after the prompt ID. Its contents are sent along with the prompt. See Context Files & Piping.
Crew
The team of robots behind a chat session. Every chat is a crew, even with one model; you add members with /add_recruit. See Crews.
Directive
A command that starts with /, such as /model, /clear or /webpage. You type directives in chat; some also work inside prompt files. aia handles them itself rather than sending them to the model. See Directives and the Directives reference.
ERB
Embedded Ruby — <%= … %> tags in a prompt file that are replaced with the result of Ruby code. Parameters and include are written this way. See ERB & Shell.
Executable prompt
A prompt file whose first line is a #! line that runs aia, so you can run the file directly like a script. See Executable Prompts.
Front matter
The YAML block between --- lines at the top of a prompt file. It holds the description, parameters, next: / pipeline: and other settings. See Prompt Files.
History file
The file that keeps the lines you type at the chat prompt, for up-arrow recall. Default ~/.prompts/_prompts.log. See Output & History.
MCP (Model Context Protocol)
An open protocol for giving models access to tools and data provided by separate programs or services. See MCP Servers.
MCP server
A program (run locally over stdio) or a web service (over Streamable HTTP) that offers tools through MCP. Configured under mcp_servers: or with --mcp. See MCP Servers.
Model
The LLM that answers, such as gpt-4o-mini (the default) or claude-sonnet-4-6. Chosen with -m. See Choosing Models.
Multi-model
Running the same prompt on several models at once, by giving -m a comma-separated list. See Multi-Model & Consensus.
Output file
The file each response is written to, as well as the terminal. Default temp.md in the current directory; change it with -o or turn it off with --no-output. See Output & History.
Parameter
A named blank in a prompt, written <%= name %> and declared in the front matter. aia asks you for a value when the parameter has no default. See Parameters.
Pipeline
A sequence of prompts run one after another in the same conversation, so each sees the responses before it. Built with -p, -n, or next: / pipeline: in front matter. See Pipelines.
Plugin
A Ruby file in your plugins directory that aia loads at startup and reloads when it changes during a chat. See Plugins.
Prompt
The text you send to the model. In aia it usually comes from a prompt file, but it can also be piped in or typed in chat.
Prompt file
A Markdown file (.md) in your prompts directory that contains a prompt, optionally with front matter, parameters, ERB and shell expansion. See Prompt Files.
Prompt ID
A prompt file's name without .md. aia code_review runs ~/.prompts/code_review.md. A prompt in a subfolder includes the folder (dev/code-review), and an ID starting with /, ./, ../ or ~/ is a path to a prompt file. See Prompt Files.
Prompts directory
Where aia looks for prompt files. Default ~/.prompts; change it with --prompts-dir or AIA_PROMPTS__DIR.
Provider
The service or local server that runs a model: openai, anthropic, gemini, ollama, lms and so on. aia --list-providers lists them. See Choosing Models.
Provider prefix
A provider name written in front of a model name, as in ollama/llama3.2, to say which provider should run it.
Recruit
A crew member you add with /add_recruit. It stays for the rest of the session and answers to @name. See Crews.
Robot
aia's name for one configured model connection — a model plus its role, tools and conversation. A chat session has at least one (the chief).
Role
A prompt file in the roles directory (~/.prompts/roles/) that tells the model who to be, such as "senior Ruby reviewer". Applied with -r. See Roles & Skills.
Shell expansion
Replacing $(command) in a prompt file with the command's output, and $NAME with an environment variable, before the prompt is sent. See ERB & Shell.
Skill
A reusable set of instructions stored as a directory containing a SKILL.md file, added to a prompt or chat with -s. See Roles & Skills.
Specialist
A one-shot robot created with /spawn to handle the next prompt only. See Crews.
System prompt
Standing instructions given to the model before the conversation. In chat, aia builds it from --system-prompt, the role and any skills.
Thinking
Reasoning that some models write inside <think>…</think> tags before answering. Hidden unless you pass --thinking.
Token
The unit models count text in — roughly three-quarters of an English word. Limits and prices are in tokens; --tokens and --cost show usage.
Tool
A function the model can call to act or look something up — read a file, search the web, query a database. Tools come from Ruby files (--tools), gems (--require) or MCP servers. See Tools.
Tool approval
The check aia applies before running tools from an MCP server marked requires_approval:. Controlled by --tool-approval. See Security.
Tool filter
Automatic narrowing of the tools offered to the model on each turn to those relevant to your prompt. On by default; --no-auto-tool-filter turns it off.
TrakFlow
A task-tracking tool aia can use to record pipelines (--track-pipeline) and plans (/plan, /task, /tasks). See Advanced Modes.

See also