WTG (What The GPT) ❓

Chat with your program logs!

A command line program that allows you to pass the terminal output of the last command run to a GPT as context for a one time question or an extended chat. Supports Unix like OSes.

Why wtg? Relevant, expresses questioning, and typeable with one hand (in fact, all the subcommands are)!

Installation

Ensure Rust is installed on your machine and then install wtg as a local binary crate

cargo install wtg

Usage

wtg supports queries (one time questions) and chats (multiple follow ups). Both can be done inside or outside a wtg session. Logs are only recorded inside wtg sessions.

Starting a wtg session

Start a wtg session in a terminal.

wtg s /path/to/log/file

Under the hood, this starts a new pseudo terminal where wtg appends commands' stdout and stderr for a log file (closely mirroring script in UNIX). Additionally, wtg adds delimiters in the log to indicate the start and end of different commands' output. This sets the WTG_LOG env variable to the provided log file.

To ask a model about the last run's output

wtg q

If no file name is provided, this implicitly queries the file specified by the WTG_LOG environment variable.

Start an extended chat about the last run's output

wtg c

This has similar logfile semantics as wtg q.

End a wtg session (a nested shell session) with the typical exit command.

Running commands outside of a session

Not all commands need to be run in a wtg session. You can run a query against any explicitly specified log file

wtg q -l /path/to/log/file

If no log file is provided, this falls back to using the file specified in WTG_LOG.

wtg c for chat can be used in a similar way.

wtg q can also take stdout/stderr from another command via pipe.

some_program 2>&1 | wtg q

Chats cannot

Additional CLI Options

Queries (wtg q) are run with a default prompt. This prompt can be customized per run

wtg q -p "A custom prompt"

The model queries and prompts can also be specified

wtg c -m "o3-mini"

Environment Variables

Environment variables are used so users can customize default behavior of wtg commands while reducing typing of redundant CLI args.

• WTG_OPENAI_KEY: Required. The OpenAI API key to use for any queries or chats.
• WTG_LOG: Optional for queries and chats. Specifies the absolute (recommended) or relative log file to use for queries and chats. If not specified, logfile arg must be provided.
• WTG_LLM: Optional. The model to use for the session (default: gpt-4o, also valid: gpt-4o-mini, o3-mini)
• WTG_PROMPT: Optional. The default prompt to use for query if none is provided by the user.

wtg queries and chats use the below environment variables. For equivalent options, the fallback order is (1) the parameter CLI argument (if applicable), (2) the environment variable, (3) the hard coded default (if applicable).

For example, the prompt used in query (not applicable to chat, since all prompts are user provided) the model will be (1) the -p parameter if provided, (2) the WTG_PROMPT env var if set, (3) the default prompt DEFAULT_QUERY.

Similarly, the log file used for contexts in queries and chats will be (1) the -l parameter if provided, (2) the WTG_LOG variable if set, (3) N/A as the log file does not have a default.

Lastly, the model used is selected by, (1) the -m if provided, (2) the WTG_LLM env var if set, (3) the default DEFAULT_LLM.

These environment variables can be added to ~/.bashrc, ~/.zshrc or similar shell configuration files.

Notes

If using wtl with tmux, it's more convenient to start the tmux session first and then start wtl. If done in the reverse order, tmux may clear the WTG_LOG env var (which the wtl session sets). You would need to reinitialize this variable or pass the logfile to the q and c subcommands.