Skip to content
llmist llmist

TUI Overview

When running llmist agent in an interactive terminal, you get access to a rich Terminal User Interface (TUI) that helps you monitor and interact with the agent's execution.

Features

Section titled “Features”

Interactive Blocks

Section titled “Interactive Blocks”

Each LLM call and gadget execution is displayed as a collapsible block:

  • LLM calls show the streaming response in real-time
  • Gadget executions show parameters, status, and results
  • Expand/collapse any block to see full details

Real-time Status Bar

Section titled “Real-time Status Bar”

The status bar at the bottom shows:

| Metric | Description | |--------|-------------| | Mode | Current mode: BROWSE, INPUT, or FOCUSED | | Profile | Active profile name (when using custom profiles) | | Tokens | Input/output tokens used (↑ input, ↓ output, ⤿ cached) | | Cost | Estimated cost in USD | | Time | Elapsed time for current operation | | Throttled | Rate limiting active: ⏸ Throttled Xs shows delay | | Retry | Retry in progress: 🔄 Retry N/M shows attempt | | Activity | Active LLM calls and gadget executions with spinner |

Rate Limiting & Retry Indicators

Section titled “Rate Limiting & Retry Indicators”

Throttling Indicator: When proactive rate limiting delays a request, you'll see:

⏸ Throttled 3s | ...

This appears when approaching configured rate limits (RPM/TPM). A conversation entry shows:

[Rate limit] Waiting 3s (RPM: 45/50, TPM: 38K/40K)

Retry Indicator: When automatic retry is active after an error, you'll see:

🔄 Retry 2/3 | ...

This shows the current attempt and total retries configured. Retry reason appears in conversation history.

Configuration:

  • Control rate limits via TOML [rate-limits] section or --rate-limit-rpm flag
  • Control retry via TOML [retry] section or --max-retries flag
  • See CLI Configuration for details

Raw Request/Response Viewer

Section titled “Raw Request/Response Viewer”

Debug LLM interactions by viewing:

  • Raw JSON request sent to the provider
  • Full response payload received

Keyboard Shortcuts

Section titled “Keyboard Shortcuts”

| Key | Action | |-----|--------| | ↑ / k | Select previous block | | ↓ / j | Select next block | | Enter | Toggle expand/collapse selected block | | r | View raw request | | R | View raw response | | Ctrl+B | Toggle browse/input mode | | Ctrl+C (×2) | Quit agent |

Browse vs Input Mode

Section titled “Browse vs Input Mode”

The TUI has two modes:

  1. Input Mode (default) - Type messages to send to the agent
  2. Browse Mode - Navigate through execution history

Toggle between modes with Ctrl+B. In browse mode, you can:

  • Navigate blocks with arrow keys or j/k
  • Expand blocks to see full content
  • Inspect raw LLM payloads

Non-Interactive Mode

Section titled “Non-Interactive Mode”

When running without a TTY (e.g., in scripts or CI), the TUI is disabled and output is streamed directly to stdout:

Terminal window
    # Runs without TUI
    echo "How many floppies for DOOM.ZIP?" | npx @llmist/cli agent -g ./floppy.ts


    # Use --quiet to suppress all output except gadget results
    npx @llmist/cli agent "Check my high score" -g ./arcade.ts --quiet