carlab-pi 0.5.10

PI terminal coding harness for the CARLab Qwen model.

PI

Installable terminal coding harness for the CARLab local Qwen model.

Install

From another machine:

scp william@128.175.213.232:/media/william/mist2/william/Github/CARLabAssistant/packages/pi-client/dist/carlab_pi-0.5.10-py3-none-any.whl .
python3 -m pip install --user ./carlab_pi-0.5.10-py3-none-any.whl

From a wheel or source directory:

python3 -m pip install carlab_pi-0.5.10-py3-none-any.whl

or:

python3 -m pip install .

Configure

Create an API key in Open WebUI, then save it locally:

pi setup --api-key <your-open-webui-api-key>

The default endpoint is:

http://128.175.213.232:8080/api/v1

Use

pi
pi --version
pi status
pi ask "Write a Python function that reads a CSV."
pi chat --file app.py
pi resume
pi resume 20260611-103000-a1b2c3d4

Inside interactive mode:

/pwd
/cd PATH
/ls [PATH]
!pytest -q
/file PATH
/exit

Use the up/down arrow keys to navigate previously sent prompts. Prompt history is stored persistently at ~/.pi/history.

PI also handles common requests without slash commands:

list the files in the current directory
read the server config file
fix the typo in readme
search the web for the latest release notes

When you ask PI to edit a file, it fuzzy-matches the file, asks the model for a unified diff, applies it, and saves the file automatically.

PI can also search the web when the harness decides the request needs current, niche, or uncertain information. Stable questions are answered without web search.

Internal searches, file reads, and tool calls stay quiet by default. PI uses their results as context and prints the final answer, not raw search pages or raw file contents. Large files are summarized into temporary code memory before they are sent to the model, so broad review/explanation requests stay inside the context window.

Configuration is stored at:

~/.pi/config.json

Older ~/.config/pi/config.json files are migrated automatically, so API keys survive package updates.

Interactive chat sessions are stored locally under:

~/.pi/<session_id>/

Each session directory contains metadata.json and transcript.jsonl. Use pi resume to choose from saved sessions by title and last prompt time, or pi resume <session_id> to resume a known session directly. When a session is resumed, PI prints the most recent five saved interactions before showing the prompt.