> Warden / blog
tui, terminal, usability

Your whole session, searchable

Warden's terminal interface is now the default. Search the whole transcript, jump to any command, and open any tool call to read exactly what the agent did.

Run Warden in a terminal now and you get the full-screen interface by default, not a scrolling line-by-line log. The reason for the switch is simple. When an agent runs for twenty minutes and makes forty tool calls, a flat log buries the one thing you need to find. The interface exists so nothing scrolls away and gets lost.

Search the transcript

Press Ctrl+F at any point and search the entire session: the model’s replies, the tool results, the system messages. It matches as you type and wraps around, so finding the place where it edited the config is two keystrokes instead of a scrollback marathon.

Jump to any command

Ctrl+P opens a command palette over every slash command Warden knows. Start typing and it fuzzy-matches, with each command’s description right there, so you don’t have to remember the exact name of the one you want. The model picker and the reasoning picker open as their own dialogs from the same place.

Open any tool call and read the whole thing

Every tool call renders as a card you can expand. It stays collapsed by default so the conversation reads cleanly, and opens when you want the detail. Inside, you get the full, untruncated output, and for a file edit you get a real diff of what changed. Nothing is clipped down to a preview. If the agent touched it, you can read exactly what it did.

Watch it work

Text streams in as the model produces it, so you read along instead of waiting for the turn to finish. A status line keeps the model, the reasoning effort, the running cost, and whether a turn is in flight all visible at once. You always know what is running and what it is costing.

Why it matters

An agent you can’t inspect is an agent you have to trust blindly. The point of all of this is the opposite: every action it took is one keystroke away, in full, after the fact. You stay the reviewer, not the spectator.

Warden is open source under the MIT license. Point it at TheAuditor for code facts, give it memory with Curator, and let Arbiter run it across providers. Try it and tell us what the interface still hides.

Was this useful?