Use mike in the terminal

Run mike with no arguments in a terminal and it opens a small full-screen app: whether mike is running and healthy, a few switches, the sessions it has captured and the requests inside them. It is for the quick look and the first setup, from an SSH session or a pane beside Claude Code. It is not a second web UI: it shows less, and everything deeper (the timeline, search, memory, steering, filters) is in the web UI.

Only a terminal gets the app. If mike is run with its output piped or redirected, it starts the server instead, as mike serve does.

Open it and leave it

1. Run mike in a terminal.
2. On a machine that has not finished setup, the app opens on the Setup screen. Otherwise it opens on Services.
3. Press ? for the keys of the screen you are on. Press Esc or ? again to close the list.
4. Press q to leave. The terminal returns to how it was.

Esc goes back one screen (from a request to its session list, from a session list to Services). It does not quit. Ctrl and / also opens the key list, which matters while you are typing into a setup field, where ? is a literal character.

mike setup opens the same app on the Setup screen, and mike setup <step> opens it at one step. See Setup.

What the screens show

The left menu lists four screens: Services, Config, Setup and Sessions. The bar at the top says whether mike is online or stopped.

Services

The first screen on a configured machine. It lists:

• Binds: each address mike listens on (the proxy, the inspect API and the web UI), each up or down.
• System: how much the store holds, mike’s memory use, and whether mike-mcp is on your PATH and can reach the inspect API.
• Capabilities: each background feature, up, degraded, inert or off.

Keys: t or ⏎ starts mike when it is stopped and stops it when it is running; r restarts it; l opens a full screen of the service’s log, with x to stop, r to restart and Tab to move between panels; s opens Setup.

Config

The switches of the Settings panel, each on or off, with the number of capabilities active at the bottom. ↑ and ↓ (or j and k) move; ⏎ flips the selected on/off switch. Settings that are not on/off, such as a text value, are shown but cannot be edited here; the screen points to /settings/workers for those. It reads and writes the same settings as the web UI’s Settings.

Setup

The first-run steps, ten in all, from “welcome” to “ready”. They are the steps of mike setup and /settings/setup, described in Setup. On the last step, ⏎ starts Claude Code through mike, so the first session is already captured. f leaves setup and keeps the answers so far.

Sessions

One line per captured session, newest first: busy or idle, the session id, the number of requests, the first model it used, the project and the session’s title. Fifty sessions show per page; PgDn and PgUp move between pages and r refreshes.

⏎ on a session lists its requests: status (complete, streaming, pending or error), request number, method, who started the turn, model and how many tools were offered. ⏎ on a request shows it in three views, cycled with Tab: its details, its request body and its response body. ↑/↓, PgUp/PgDn scroll, g returns to the top, r refreshes.

The sessions here are the ones mike captured, not only those started with mike spawn. The docs session of the example would appear here once it has made a request.

Keys

The menu on the left has focus first, so ↑ and ↓ there change the screen. Press ⏎ or Tab to move into the screen, Shift+Tab to return to the menu. The terminal app has no mouse support.

What it leaves to the web UI

The terminal app has no timeline, no search, no memory or steering, no filters, no projects and no messages between sessions. The request bodies are shown as stored, one scrolling view each, without the conversation laid out. For any of those open the web UI at http://localhost:9990/, which the setup’s last step also offers with b.

When it shows nothing or fails