Print mode & scripting

Run Tau non-interactively for a single prompt — ideal for scripts, pipes, and CI.

Print mode runs a single prompt without the interactive UI and writes the result to the terminal. It’s the right choice for scripts, pipelines, and one-off questions.

Basic use

tau -p "summarize the changes in the last commit"

The -p / --print flag switches Tau into print mode; the prompt itself is a plain positional argument, the same as an initial prompt for the TUI. Print mode still uses the full coding-session environment — the same tools, project context, and session storage as the TUI — so its turns are saved under ~/.tau/sessions/ too.

Put flags before the prompt. Tau accepts multi-word prompts without quoting, so anything after the last recognized flag — including tokens that look like other flags — is treated as prompt text:

tau -p --mode json "list the public functions in src/app.py"

Output formats

Choose how results are written with --mode. Passing --mode on its own also switches Tau into non-interactive mode, so -p is optional once --mode is set:

tau -p --mode text "list the public functions in src/app.py"        # default, human-readable
tau --mode json "list the public functions in src/app.py"           # JSON, for parsing
tau -p --mode transcript "list the public functions in src/app.py"  # structured transcript

• text — plain text with ANSI styling, for reading.
• json — machine-readable, for piping into other tools.
• transcript — a structured record of the turn.

Piped stdin is merged into the prompt, so you can feed file contents in:

cat README.md | tau -p "Summarize this text"

A piped body can also be the entire prompt — tau -p with no positional text and piped stdin is valid:

cat README.md | tau -p

Choosing provider, model, and directory

The same selection flags work in print mode:

tau -m gpt-5.5 -p "explain this module"
tau --provider local -p "explain this module"
tau --cwd ./services/api -p "audit for secrets"

Resume a conversation

Pass an existing session id to run a follow-up turn non-interactively:

tau --print --session <session-id> "Follow-up message"

Tau appends the turn to the existing transcript and sends its active conversation history to the model. It also uses the session’s saved working directory, provider, and model unless explicit selection flags override them. This works with every output mode and keeps stdout dedicated to that mode. Unknown ids fail without creating a session. --session cannot be combined with --new-session or --session-id.

Use tau sessions to find session ids.

Recording the session id

Automation can choose the exact id of a new print-mode session with --session-id. This keeps stdout dedicated to the selected output format and avoids scanning ~/.tau/sessions/:

worker_session_id="$(python -c 'import uuid; print(uuid.uuid4().hex)')"
tau --print --new-session \
  --session-id "$worker_session_id" \
  --cwd /path/to/project \
  "review the current changes"
printf 'Tau session: %s\n' "$worker_session_id"

Ids may contain letters, numbers, ., _, and -, must start and end with a letter or number, and may be at most 128 bytes. default and index are reserved. Use a unique id for each worker. Tau atomically reserves the transcript and exits with an error rather than opening or overwriting an existing session, even if an unindexed transcript already uses that id or two workers start at the same time. The option applies to text, JSON, and transcript modes without adding metadata to their stdout output.

Exit status

Print mode exits non-zero if the run fails, so you can use it in scripts:

if tau --mode text -p "do the tests pass? answer yes or no" | grep -qi yes; then
  echo "looks good"
fi