TUI Dashboard

The 2Signal TUI is an interactive terminal dashboard for real-time monitoring of your AI agents. Think lazydocker, but for AI traces.

Installation

pip install twosignal[tui]

Requirements

Launch

twosignal-tui

# or run as a module
python -m twosignal_tui

On first launch, you'll be prompted for your email and password. If you've already run twosignal login via the CLI, you'll be authenticated automatically.

Keyboard Shortcuts

The TUI is fully keyboard-driven. Shortcuts are shown in the footer bar, but here is the full reference:

Screens

Dashboard Screen

The dashboard is the main screen you see after selecting a project. It provides a real-time overview of your agent's health:

• Trace count sparkline — a compact chart showing trace volume over the last hour. Spikes indicate bursts of activity; gaps may indicate downtime.
• Error rate — the percentage of traces with ERROR status, updated in real time.
• Average latency — mean trace duration across recent traces, displayed with a trend indicator.
• Recent traces table — the most recent traces with their status, duration, cost, and timestamp. Select a row and press Enter to drill into the trace detail.
• Activity log — a scrolling feed of recent events: new traces, evaluator results, errors, and configuration changes.

The dashboard auto-refreshes every few seconds. Press r to force an immediate refresh.

Trace Detail View

When you select a trace, the detail view shows a hierarchical span tree:

• Span tree — traces are broken into nested spans displayed as an indented tree. Each span shows its name, type, duration, and token count. Parent-child relationships are represented with box-drawing characters.
• Input/output inspection — select a span and press Enter to view its full input and output. The JSON viewer provides syntax highlighting and collapsible sections for large payloads.
• Scores — if evaluators have run on this trace, scores are displayed alongside the trace summary. Each score shows the evaluator name, numeric score, and pass/fail status.
• Metadata — view tags, model names, token usage, and cost for each span.

Evaluator Management

The Evaluators screen lets you create, edit, and delete evaluators without leaving the terminal. When you create a new evaluator:

1. Select the evaluator type from a list (CONTAINS, REGEX_MATCH, JSON_SCHEMA, LATENCY, COST, SIMILARITY, LLM_JUDGE).
2. Fill in the configuration form. Fields are validated in real time — the TUI shows errors inline if a field is invalid.
3. Give the evaluator a name and confirm. It takes effect on the next incoming trace.

You can also toggle evaluators on/off and delete them from the list view. Changes are saved immediately to the server.

Features

• Live updates — auto-refreshing dashboard with sparkline charts
• Span tree — hierarchical view of trace spans with timing and tokens
• JSON viewer — inspect span inputs and outputs with syntax highlighting
• Keyboard-driven — full navigation via keyboard shortcuts (shown in footer)
• Shared auth — same session as the CLI

Configuration

# connect to a local dev instance
TWOSIGNAL_URL=http://localhost:3000 twosignal-tui

The TUI reads from TWOSIGNAL_URL env var, ~/.twosignal/config.json, or defaults to https://app.twosignal.dev.

CLI vs TUI

Both tools share the same authentication and can do many of the same things. Choose based on your workflow:

Troubleshooting