Output, History and Usage¶

Where AIA's answers go, what it remembers between sessions, how to see token counts and costs, and how to turn on logging when something goes wrong.

The output file¶

Besides printing each answer, AIA appends it to an output file. Out of the box that file is temp.md in your current directory.

At the start of each run, AIA empties the output file (unless you ask it to append, below).
Each answer is written after an AI: label. In chat mode, each of your messages is written too, after You:, so the file becomes a transcript.
Token and cost tables from --tokens / --cost, and the CSV from /cost, are written there as well.

temp.md after a short chat

AI: **Tobor** (2.4s):
Here's a summary of the report...

You: Which section has the most risk?

AI: Section 4, the vendor migration...

Choosing where output goes¶

You want…	Do this
A specific file	`--output notes/answer.md` (`-o`). Relative paths are relative to where you run AIA.
The default `temp.md`	`-o` with no file name. Put it last or before another option: `aia -o summarize notes.txt` takes `summarize` as the output file name
No output file at all	`--no-output`
To add to the file instead of replacing it	`--append` (`-a`); `--no-append` to switch it back off
No output file, ever	`output.file: ~` in your config file
To always append	`output.append: true` in your config file

# ~/.config/aia/aia.yml
output:
  file: ~          # stop writing ./temp.md

Keep a running journal

aia --chat -a -o ~/journal/ai.md appends every chat to the same file, giving you a searchable log of everything you've asked.

When you generate an image with --paint, the image is saved next to the output file with a .png extension (temp.md becomes temp.png).

Chat input history¶

AIA remembers the lines you type in chat, so Up and Down bring back earlier messages, even from previous sessions. It keeps the last 50 entries.

Setting	Default
`output.history_file`	`~/.prompts/_prompts.log`

aia --chat --history-file ~/.config/aia/work_history    # use another file
aia --chat --no-history-file                            # don't save input history

If the history file grows beyond 10 MB, AIA renames it to <file>.1 at the start of a session and starts a fresh one, printing History log rotated: ….

Note

The history file holds what you typed, not the model's answers. For a record of the whole conversation, use the output file.

Token counts and time: `--tokens`¶

aia --tokens summarize report.txt

After each answer AIA shows how many tokens went in and came out, and how long the answer took:

┌─────────────┬───────┬────────┬───────┬──────┐
│ Model       │ Input │ Output │ Total │ Time │
├─────────────┼───────┼────────┼───────┼──────┤
│ gpt-4o-mini │  1843 │    212 │  2055 │ 2.3s │
└─────────────┴───────┴────────┴───────┴──────┘

That compact table is what you see after each chat turn. When you run a prompt without --chat, the same figures appear under a Multi-Model Token Usage heading with a TOTAL row, even for one model, because one-shot runs go through AIA's robot network.

Input tokens include everything sent: the role, the prompt, skills, piped text, context files and, in chat, the conversation so far. If a request is bigger than you expected, this is where you'll see it.

Cost: `--cost`¶

--cost does everything --tokens does and adds what the answer cost, plus what 1,000 answers like it would cost:

┌─────────────┬───────┬────────┬───────┬──────────┬───────┬──────┐
│ Model       │ Input │ Output │ Total │ Cost     │ x1000 │ Time │
├─────────────┼───────┼────────┼───────┼──────────┼───────┼──────┤
│ gpt-4o-mini │  1843 │    212 │  2055 │ $0.00040 │ $0.40 │ 2.3s │
└─────────────┴───────┴────────┴───────┴──────────┴───────┴──────┘

Prices come from the RubyLLM model catalog. When a model has no price listed (local models, very new models), the cost shows N/A.

With several models, you get one row per model under a Multi-Model Token Usage heading, plus a TOTAL row. Time in the total row is the slowest model's, because the models run in parallel.

A whole session as CSV: `/cost`¶

In a chat, /cost prints one CSV row per turn so far, plus a totals row, with the columns model,input_tokens,output_tokens,total_tokens,cost,elapsed. The CSV is also appended to the output file, so it's easy to pull into a spreadsheet.

Logging and debugging¶

AIA has three logs:

Log	Records
`aia`	AIA's own activity, including the full configuration at startup (at debug level)
`llm`	Requests to and responses from model providers
`mcp`	Connections to MCP servers

By default all three go to standard output and only show warnings and errors.

Option	Effect
`--log-level LEVEL`	Set every log to `debug`, `info`, `warn`, `error` or `fatal`
`-d`, `--debug`	Shortcut for `--log-level debug`
`--no-debug`	Turn debug off again (handy when your config file turns it on)
`--log-to FILE`	Send all three logs to `FILE` instead of the terminal
`-v`, `--verbose`	Show extra detail, such as which tools were picked for each chat turn

aia --debug --log-to /tmp/aia.log summarize report.txt
tail -f /tmp/aia.log            # in another terminal

Log files roll over daily. To set each log separately, use the logger section of the config file:

# ~/.config/aia/aia.yml
logger:
  aia:
    file: ~/.config/aia/logs/aia.log
    level: info
  llm:
    file: ~/.config/aia/logs/llm.log
    level: debug
  mcp:
    file: STDERR
    level: warn

file can be a path, STDOUT or STDERR. --log-level and --log-to override these settings for one run. In a chat you can change log levels on the fly with /config, for example /config logger.llm.level debug.

Saving your settings¶

--dump FILE writes the configuration AIA would use (after your config file, environment variables and command-line options are combined) to a YAML file, then exits:

aia --model claude-sonnet-4-5 -t 0.3 --dump current.yml

That's a quick way to check what a combination of settings resolves to, or to start a new config file. The file must end in .yml or .yaml. See Config File.