Betty

A real-time TUI monitor for Claude Code sessions.

Install

curl -fsSL https://betty4.sh/install.sh | bash

Or directly with uv / pip:

uvx --from betty-cli betty  # run without installing
uv tool install betty-cli   # install permanently
pip install betty-cli       # with pip

Use

# Start betty
betty

# In another terminal, run Claude Code as usual
claude

The companion auto-detects your session. No hooks or configuration needed.

Options

Flag	Description
--global, -g	Watch all projects
--worktree, -w	Watch across git worktrees
--style	UI style (rich or claude-code)
--version, -v	Show version

Commands

Command	Description
config	Configure LLM summarization and UI settings
mock --demo	Generate mock sessions for development

Keybindings

Navigation

Key	Action
j/k	Navigate turns
g/G	Jump to beginning/end
1-9	Switch sessions
h/l	Switch panels (manager expand mode)

Display

Key	Action
o / Space / Enter	Expand/collapse turn or span
e/c	Expand/collapse all
f	Cycle filters (All, Spans, Tools, Read, Write, Edit, Bash)
s/S	Toggle summaries / Summarize all

Views

Key	Action
M	Toggle manager view
T	Toggle tasks view
P	Toggle plan view
I	Toggle insights (analysis) panel

Analysis & Annotations

Key	Action
A	Analyze selected turn/span/session
[/]	Zoom analysis level (turn / span / session)
n	Annotate selected turn
a	Toggle/clear alerts

Agent

Key	Action
B	Toggle agent panel (closed / full / compact)
?	Ask Betty a question about the session

Other

Key	Action
O	Open PR in browser
x	Export to Markdown
m	Edit monitor instructions
D	Delete session
Esc	Close panel / clear selection
q	Quit

Betty Agent

Betty Agent is a continuous session observer that tracks what Claude Code is doing and flags problems in real time. It combines heuristic detectors with optional LLM-powered narrative and drift detection.

Enable

betty config --agent

Or set the environment variable BETTY_AGENT_ENABLED=true.

What it does

• Goal tracking — extracts the session goal and current objective, updating as the user gives new instructions
• Progress assessment — classifies sessions as on_track, stalled, or spinning using error rates, retry patterns, and tool diversity
• Error spike detection — warns when error rate exceeds 40% in recent tool calls
• Retry loop detection — flags when the same tool is called 3+ times consecutively
• Stall detection — notices gaps of 2+ minutes between turns
• File change tracking — logs Read/Write/Edit operations with line counts
• Milestones — marks every 10th tool call and 5th user message
• LLM narrative (optional) — generates a 2-3 sentence situation report describing current activity
• Goal drift detection (optional) — compares recent activity against the session goal and warns if the assistant has gone off track
• Ask Betty — press ? to ask a natural-language question about the session; Betty answers citing turn numbers and file paths

Configuration

The agent uses your existing LLM configuration (set via betty config). LLM features (narrative, drift detection, goal determination, Ask Betty) require a configured LLM provider. Heuristic detectors work without one.

Setting	Default	Description
enabled	false	Enable the agent (opt-in)
update_interval	5	Minimum turns between LLM updates
max_observations	50	Max observations kept per session

Observations and reports are cached to disk (~/.cache/betty/) and persist across restarts.

License

MIT