AI agents

A terminal that works with your agents.

Konsoll recognizes AI coding agents, shows what they are doing in the tab, and brings their questions to you, wherever you are.

The tab shows the agent's state

Connect an agent once and it reports its state from its own hooks. With several panes in a tab, the one that most needs you sets the icon.

  • Workingon a turn
  • Waitingfor permission or an answer
  • Donewith the turn
  • Failed
An agent asking for permission to run npm test in the webapp tab, other tabs with agents working and done

Allow or Deny, from the notification

When an agent asks to run a tool, the macOS notification shows what it wants — the command, the file — with Allow and Deny. The answer goes to the agent's pane as if typed there, and Konsoll stays in the background.

Notifications also tell you when an agent waits for an answer, finishes (with the task it was given) or fails while you look elsewhere. The Dock icon counts the agents waiting for you.

The agent asks to run npm test, the request is allowed from Konsoll and the agent finishes the turn

Every agent, in one list

Window ▸ AI Agents (⇧⌘A) lists every pane running an agent in every window, the ones that need you first: what each asks for, the task it works on and how long it has waited, with Allow, Deny and Show.

Window AI Agents: two agents asking for permission with Allow and Deny, one working on a task, one done

Your terminal as the agent's context

Right-click a pane: Send Selection to Agent and Send Last Command to Agent paste the text, or a command with its exit status and output, into the agent's prompt, ready for your question.

Shell ▸ New Agent in Worktree… creates a git worktree on a branch and starts an agent there in a new tab, so agents work in parallel apart from your checkout.

A failed npm test, and the pane menu's Send Last Command To submenu listing three agents

Over ssh, with nothing to install

ssh host in a zsh or bash tab brings Konsoll's shell integration to the host for that session. Claude Code, Codex and Qwen Code started there show their state in the tab and ask through Konsoll's notifications. Everything lives in a private temporary folder removed when the shell exits.

ssh to a Linux host, Claude Code started there asks to run a command, and the request is allowed from Konsoll

Tools for agents (MCP)

konsoll --mcp gives an agent running in Konsoll tools that act from its own pane.

ToolWhat it does
open_paneOpens a pane beside the agent's running a dev server or watcher, without taking the focus.
run_in_backgroundStarts a Background Command.
background_job_statusThe state and last output of a background command that agent started.
recent_commandsYour last commands in the agent's pane with their status and output: “why did my build fail?” works without pasting.
list_panesThe panes of the window: tab, program, directory — no output.
read_paneReads another pane only after you choose Allow Once.

Connect an agent

Start an agent in a pane and Konsoll offers to connect it. Or open Settings ▸ AI Agents: Connect adds Konsoll's hooks to the agent's settings (keeping a backup), Remove takes them out, and Copy gives the MCP setup for each agent. The hooks do nothing outside Konsoll.

Claude Code can use the Konsoll plugin instead:

claude plugin marketplace add rask-project/Konsoll
claude plugin install konsoll@konsoll
Settings AI Agents: Claude Code and Qwen Code connected, Codex not connected, and the MCP setup