Quick Tour¶
A ten-minute walk through the things you'll do with AIA most often. Each stop links to the chapter that covers it in depth. You only need AIA installed and one API key (see Installation).
Follow along without clutter
The examples pass --no-output so AIA doesn't also write each answer to ./temp.md. If you've already set output.file: ~ in your config file, you can leave it off.
1. Ask a question¶
The fastest way to use AIA is to pipe a question into it:
AI:
**Tobor** (1.2s):
`tee` reads standard input and writes it both to standard output and to one or more files.
When there's no prompt ID on the command line, whatever arrives on stdin is the prompt.
2. Save a prompt and run it by name¶
Prompts you use more than once belong in files. Create one in your prompts directory:
cat > ~/.prompts/explain.md <<'PROMPT'
Explain the following to a developer who is new to it.
Be brief, and give one concrete example.
PROMPT
The file name, without .md, is the prompt ID:
Now the prompt comes from the file and stdin is appended to it. See Running Prompts.
3. Give it files to work on¶
Any files listed after the prompt ID are added to the prompt as context:
You can list as many files as you want. See Context Files and Piping.
4. Fill in parameters¶
A prompt can declare parameters in YAML front matter and use them with ERB:
---
parameters:
language: Spanish
---
Translate the following text into <%= language %>. Keep the formatting.
Press Enter to accept the default shown in parentheses. See Parameters.
5. Pull in live data with the shell¶
$(command) in a prompt is replaced by that command's output before the prompt is sent:
6. Chat¶
Add --chat to keep the conversation going after the first answer, or start a chat with no prompt at all:
Entering interactive chat mode...
════════════════════════════════════════════════════════════
Follow up (cntl-D or 'exit' to end) #=>
What's a good name for a CLI that converts CSV to JSON?
Inside a chat, lines that start with / are directives: /model, /include notes.md, /checkpoint, /help. See Chat Sessions.
7. Pick a model¶
aia --no-output --model claude-sonnet-4-5 explain lib/parser.rb
aia --available-models claude # search the model catalog
See Choosing Models.
8. Ask several models at once¶
Each model answers in parallel. Add --consensus to have their answers merged into one. See Multiple Models and Consensus.
9. Put on a role¶
A role is a reusable persona stored in ~/.prompts/roles/:
cat > ~/.prompts/roles/reviewer.md <<'ROLE'
You are a meticulous senior code reviewer. Point out bugs first, then style.
ROLE
aia --no-output --role reviewer explain lib/parser.rb
See Roles and Skills.
10. See what it costs¶
After the response AIA prints a table of input/output tokens, elapsed time and cost. See Output, History and Usage.
Where to go from here¶
You've now seen most of what AIA does day to day. Two good next steps:
- Your First Prompt: build a real, reusable prompt step by step
- Running Prompts: every way of telling AIA what to run
Later, when you want more: Pipelines, Tools, MCP Servers.