Ask Jimothy
Ask Jimothy is a chat assistant built into the app. Click the raccoon in the bottom-right corner, or press ⌘J (Ctrl+J on Windows and Linux), and ask in plain language:
- “Why did my last run fail?”
- “Retry the failed step with a clearer prompt.”
- “What is waiting for my approval?”
- “Set up a pipeline that fixes GitHub issues labeled bug, and a trigger for it.”
Jimothy can see and change everything the app can: runs and their logs, projects, pipelines, triggers, boards, accounts, harnesses, notifications and settings. It reads before it answers, so an explanation of a failure quotes the lines from the step’s log that matter. When it fixes something (retrying a run, editing a step, saving a pipeline), it says what it changed.
Each message tells Jimothy which page you’re on, so “why did this run fail?” works from a run, and “add a review step to this pipeline” works from the pipeline editor.
Chat about one project
Section titled “Chat about one project”Under the message box, the project picker (it says All projects at first) narrows the chat to one project. Each message then tells Jimothy which project you mean and which pipelines are in it, so “what failed this week?” means that project’s runs, and a new pipeline, trigger or board lands in that project. Pick All projects to work across everything again. The choice applies from your next message, and each chat remembers its own project; the history list shows it.
It runs on your harnesses
Section titled “It runs on your harnesses”Jimothy doesn’t need its own API key. It uses one of the harnesses you already have, with your existing subscription or key:
| Harness | How Jimothy uses it |
|---|---|
| Claude Code | The default when it’s installed. Each message runs claude with only Jimothy’s tools: no file editing, no shell, and none of your other MCP servers. Later messages resume the same session, so it remembers the conversation. |
| Codex | Runs codex exec read-only, with Jimothy’s tools added. Later messages resume the same thread. |
| Claude API and OpenAI-compatible APIs | Jimothy calls the model directly and runs its tools inside the app. |
Click the sliders button in the chat header to pick the harness and model. The choice is saved, and the header shows what the next message will use. Other CLIs (Gemini, OpenCode, Cursor and so on) aren’t offered yet.
Claude Code and Codex reach Jimothy’s tools through a private endpoint on 127.0.0.1 with its own token. It exists only while the app runs, and works whether or not the MCP server for outside clients is turned on.
Confirmations
Section titled “Confirmations”Reading and normal changes happen right away. Before anything that can’t be undone (deleting a run, pipeline, project, trigger, board, account, harness or notification channel; cancelling a run; resetting a trigger), Jimothy stops and shows a card in the chat. Allow lets it go ahead. Don’t tells Jimothy you declined, and it won’t try again unless you ask.
Like other AI clients, the assistant can’t change remote access or the MCP server settings.
While it works
Section titled “While it works”Each step Jimothy takes shows as a small chip under its reply, such as get run or get run logs with the ids it used, plus a check, a warning, or a stop sign for a declined action. Stop (the square button) ends the current reply. New chat (the pencil button) starts over.
While a reply is running, a little Jimothy runs along the bottom of the window, and now and then does a somersault, stands up to dance, or waves. To turn him off, switch off Show Jimothy running around while he works in the chat settings. He also stays hidden when your system is set to reduce motion.
Pop-up or side panel
Section titled “Pop-up or side panel”The placement button in the chat header (its icon shows where the chat is now, with a small arrow) opens a menu to move the chat. The same choice is under Show the chat in the chat settings (the sliders button):
| Choice | What you get |
|---|---|
| Pop-up | The default. The chat opens over the page from the raccoon button in the bottom-right corner. |
| Side panel | The chat is docked on the right of the window, next to the page. Drag its left edge to make it wider or narrower (double-click the edge to reset). Collapse (the panel button in its header) shrinks it to a slim bar with Jimothy’s head; click the head to open it again. Jimothy remembers whether it’s open and how wide it is. On a narrow window or a phone it falls back to the pop-up. |
| On the desktop | Jimothy floats outside the app window (see below). |
⌘J opens and closes the chat wherever it lives. Everything on this page is also in the ⌘K command palette: type jimothy to see it, or search your saved chats by title.
Previous chats
Section titled “Previous chats”Every conversation is saved after each reply. Click the history button (the clock) in the chat header to see your previous chats, newest first, with when you last used each one and how many messages it has.
- Go back to a chat: click it. Its messages come back and you can keep going. Claude Code and Codex resume the same session, so Jimothy remembers everything from before, even after you restart the app. A chat you started on another harness continues on the current one, which gets the earlier conversation as context.
- Rename: hover a chat and click the pencil. Type a name and press Enter (Esc cancels). Renamed chats keep your name.
- Delete: click the trash icon, then Delete to confirm.
- New chat (the pencil-on-paper button, or New chat in the history list) starts fresh. The chat you were in stays in the history.
You can’t switch chats while Jimothy is answering; stop the reply or wait for it.
Float it on the desktop
Section titled “Float it on the desktop”Jimothy can leave the app window and float on top of your other apps, so you can ask it things while you work somewhere else.
| To | Do this |
|---|---|
| Undock | Drag the corner button out of the app window (the head appears where you let go), choose On the desktop from the placement button in the chat header, switch on Float Jimothy on the desktop, outside this window in the chat settings, or choose Float Jimothy on the desktop in the tray menu. |
| Move it | Drag the head anywhere, on any display. It stays where you leave it, also after a restart. |
| Chat | Click the head. The chat opens beside it, on whichever side has room, and stays on top until you close it (×, Esc, or click the head again). |
| Dock | Drag the head onto the app window’s bottom-right corner (it fades when it’s over the spot), choose Pop-up or Side panel from the placement button in the floating chat’s header, or right-click the head and choose Dock in the app window. |
While it floats, ⌘J in the app window opens the floating chat, and messages you send from it use the page the app window shows as context. The head is always on top and appears on every desktop (Space). Right-click it for Open Jimothy and Dock in the app window. Floating is only available in the desktop app, not over remote access.
Voice input
Section titled “Voice input”Click the microphone next to Send and talk. The button turns red and pulses with your voice. Click ■ to stop, and the text appears in the message box so you can check it before sending. Esc cancels a recording. The first time, your system asks for microphone permission.
By default, speech is turned into text on your computer with OpenAI’s Whisper model, run by Transformers.js (on the GPU through WebGPU when it’s available). The model downloads from Hugging Face the first time you use the mic, with progress shown in the message box, and is cached after that. No audio leaves your machine.
Pick a different model under Voice input in the chat settings:
| Choice | Notes |
|---|---|
| Whisper Tiny | English only, fastest, about 40 MB. |
| Whisper Base | The default. Many languages, about 80 MB. |
| Whisper Small | Many languages, most accurate, about 250 MB. |
| Another Hugging Face model… | Any Transformers.js speech recognition model id, e.g. onnx-community/whisper-large-v3-turbo. |
| Transcription API | One of your OpenAI-compatible API harnesses (OpenAI, Groq, a local Whisper server). Audio is sent to <base URL>/audio/transcriptions with the harness’s key, using the model you name (default whisper-1). |
Troubleshooting
Section titled “Troubleshooting”- “No harness can run Jimothy”: install Claude Code or Codex, or set up an API harness, then pick it in the chat settings.
- “Harness isn’t available”: Jimothy couldn’t find the CLI. Check Settings → Harnesses and the extra PATH entries in Settings → General.
- Codex says a model isn’t supported: ChatGPT accounts only allow some Codex models. Pick another model in the chat settings.
- The microphone doesn’t work: allow Jimothy under System Settings → Privacy & Security → Microphone (macOS), then try again.