The UI as a Source of Context

Introduction

New Hermes users see a chat box. Experienced Hermes users see a dashboard. The difference is not the screen, it is the habit of reading it. Every time you launch hermes, the terminal tells you which model is active, which tools are loaded, how much context you have used, what each tool call cost, and how long you have been running. None of this requires a question to the agent. It is all on screen.

Key Concepts

• Welcome banner: A one-time snapshot of the runtime, printed at session start.
• Status bar: A live one-line meter that updates as you work.
• Conversation stream: The scrolling area where your messages and the agent's responses appear.
• Input prompt: The persistent text field at the bottom where you type.

Real World Context

When a junior dev asks "which model are you using right now?" they have not yet learned to read the banner. When a senior dev asks the same question, something has gone wrong (the banner does not match what they remember setting). The UI is the cheap, always-on source of truth. Reading it well saves dozens of small interruptions per day.

Deep Dive

The Hermes CLI is built around four fixed zones, each with a single job. Reading top-to-bottom they are:

text
┌─────────────────────────────────────────────────────────┐
│  Welcome banner: model · backend · cwd · tools · skills │  ← printed once at launch
├─────────────────────────────────────────────────────────┤
│                                                         │
│  Conversation stream: messages, tool calls, results     │  ← scrolls as you work
│                                                         │
├─────────────────────────────────────────────────────────┤
│  Status bar: model · tokens · context% · cost · time    │  ← updates in real time
├─────────────────────────────────────────────────────────┤
│  Input prompt: where you type the next message          │  ← always at the bottom
└─────────────────────────────────────────────────────────┘

The banner answers "what is this session?". You read it once at launch and refer back if you suspect something changed (e.g., after a /model switch).

The conversation stream answers "what has happened so far?". It is your transcript, including the actions the agent took on your behalf.

The status bar answers "where am I right now?". It is a real-time meter for context fill, cost, and elapsed time.

The input prompt answers "what do I want to do next?". It is always at the bottom, even while the agent is streaming above.

These four zones are stable across every Hermes surface that has a UI, including resumed sessions. Once you learn them, you stop hunting for information and start glancing for it.

Common Pitfalls

1. Treating the UI like noise: Beginners ignore the banner and status bar, then ask the agent "which model are you?" wasting tokens for an answer that was already on screen.
2. Reading only the conversation stream: The stream tells you what happened in chat, but cost, context, and elapsed time live in the status bar.

Best Practices

1. Read top-down on every launch: Banner first, then check the status bar before you type. It takes two seconds and prevents misalignment for the rest of the session.
2. Treat the status bar like a car dashboard: A casual glance every few turns is enough. You are watching for trends (rising context, climbing cost), not exact numbers.

Summary

• The Hermes CLI has four zones: banner, conversation stream, status bar, input prompt.
• Each zone answers a different question about the session.
• Reading the UI is faster and cheaper than asking the agent.
• Habitual reading of the dashboard is the first marker of CLI fluency.

# Launch Hermes and immediately read the dashboard
hermes
#
# 1. Banner appears: model, terminal backend, cwd, tools, skills
# 2. Status bar shows: model | tokens | context% | cost | elapsed
# 3. Input prompt waits at the bottom
#
# Now you can type with full awareness of where you are.