weavecode

Local JSONL logging and terminal UI for Weights & Biases Weave traces — with git snapshotting so you can replay exactly what code ran at each trace.

Install

pip install weavecode

Usage

import weave
import weavecode

# 1. Init weave as normal, then wrap with weavecode
client = weave.init("my-project")
weavecode.init(client, git_path="/path/to/your/repo")

# 2. Decorate functions you want to trace
@weave.op
def my_function(prompt: str) -> str:
    ...

# 3. Run your code
my_function("hello")

Every call is logged locally to ~/.cache/codeweave/<project>/ as a JSONL file, with a git snapshot of your repo at the time of the run.

View traces

weavecode
# or pass a project name directly:
weavecode my-project

This opens a terminal UI with:

• Runs list — all logged runs for the selected project, with 📸 if a git snapshot exists
• Call tree — the full trace tree for the selected run
• Source viewer — the exact source code at snapshot time, with the traced function and callsite highlighted
• Detail panel — inputs, outputs, errors, git info, and a clickable W&B link

Keybindings

Git snapshotting

When you pass git_path, weavecode commits a snapshot of your repo at the start of each run (stashing any dirty changes first). The snapshot SHA is stored in each log entry. The source viewer uses git show <sha>:<file> to display the code — it never modifies your working tree.

Requirements

• Python >= 3.8
• textual >= 0.50.0
• weave (Weights & Biases)