Skip to main content

Chat

Chat is where you work with your agents. It's a fast, live web interface — a window into the agent's work as it unfolds.

Watch the work unfold​

When an agent works, everything streams to your screen in real time:

  • Text and reasoning as the agent thinks through the problem.
  • Tool calls as it runs them — searching, editing, fetching, querying.
  • File edits as they're made.
  • Plans and to-do lists that update live as the agent checks items off.

Long runs of tool activity fold into a quiet one-line summary ("N steps", or a live "Working · …" while the agent is busy) that expands on click to the full detail. Questions, approvals, plans, and file outputs always stay visible. Prefer the classic always-expanded view? Switch Chat activity to Detailed under User Settings → General → Appearance.

note

Plans, live to-do checklists, and sub-agents are capabilities of the Claude Code and Codex engines. The lighter direct engine streams text and tool calls but doesn't show plans or to-dos. See AI Engines.

Writing a message​

Enter adds a new line in the message box. To send, press Shift+Enter or Ctrl+Enter on a desktop, or click or tap the send icon.

Attach files and images​

Drop a file or image right into the chat — photos, PDFs, Office documents, text, data files, audio, and video are all supported. Drag files onto the composer, paste them with Ctrl/Cmd+V, or pick from the file browser. (HTML and SVG uploads are blocked for safety.)

Each attachment chip shows real upload progress. Big batches queue automatically, files too large for the install's limit say so on the chip, and a failed upload explains why with a one-tap retry. Large files upload in resumable parts, so a 300 MB video survives a CDN's request limits and a brief network blip. When the agent's workspace syncs to remote machines, the chip keeps reporting until the file has actually arrived everywhere. Tap an attached photo to inspect it full-screen before sending.

Rich results inline​

Agents don't just reply with text. Right in the conversation you'll see:

  • Image galleries for pictures an agent produces or shows.
  • Interactive artifacts — charts, tables, calculators, and small tools the agent builds as real, live UI. See Artifacts.
  • Documents for the Word, Excel, PowerPoint, and PDF files the agent delivers, which open beside the chat.
  • Media players for audio and video, with full playback controls.
  • Downloadable files as clickable cards.
  • Clickable file paths — when a reply mentions a file in the agent's workspace, the path becomes a chip that opens a live preview (documents, images, media) with a download button — including files on connected remote machines.

Documents beside the chat​

When the agent delivers a Word, Excel, PowerPoint or PDF file, it opens in a window beside the chat, or floating over it on a phone. Each document gets a tab, Versions lists every version delivered in the chat, and you can edit the live file in place when your role allows it. Your edits save on their own. Files on a remote machine work the same way.

A live terminal, when you want it​

For Claude Code and Codex agents, a chat can run as an interactive terminal embedded right in the dashboard. You watch the agent's native interface live and can type into it yourself — the full command-line experience, wrapped in OtoDock's files, tools, and security. Switch a chat into terminal mode (or back) whenever it helps. See Live terminal.

Approvals in the flow​

Depending on the permission mode, an agent pauses before sensitive actions — editing files, running certain commands, using a tool — and asks you right in the chat. Click Allow or Deny and it continues. Switch modes any time — from approving each sensitive step to fully autonomous. Claude Code and Codex chats ask in the same way, before any command or edit the mode doesn't allow on its own. Claude Code also offers a read-only Plan mode for investigate-first work.

Don't wait your turn​

While the agent is working, just send a new message to add a detail or change its course. On Claude Code and Codex, your message reaches the agent at its next tool step, so nothing stops and the work carries on with your input folded in. This works in the sandbox, on a paired remote machine, and in a delegated worker that is still busy.

When the agent can't take a message mid-turn, it waits in the chat, marked queued, and goes out as soon as it can. The x on a queued message takes it back into the message box.

Switch model mid-chat​

Change the model from the dropdown without losing your place — try a question on a faster model, then switch to a more capable one for the heavy lifting, mid-conversation. While the chat's session is live you switch among the models of its current engine. Once the session has ended, the agent's other engines appear in the picker too: pick one and the chat restarts on the new engine from its saved history, after a confirmation.

Work across many conversations​

  • Start a new chat any time, with any agent you have access to.
  • Resume past conversations from the sidebar, grouped by recency — they pick up right where you left off, even after a server restart. Rename or delete them from the same place.
  • Search across all your chats, or within one chat, to find that thing you discussed.
  • See what's running — an Active now section above the chat history and on the agent's home page shows every session generating right now. Both panels collapse and remember your choice.
  • Up to eight chats running at once. A chat keeps a live agent session while you use it, and an idle one closes by itself after a while. Each person can have eight running at a time: opening a ninth first closes one of your chats that has been idle for a few minutes, and if none has, you're asked to end one. This doesn't limit your chat history, and tasks, phone calls and meetings don't count. An admin can change the number with OTODOCK_MAX_SESSIONS_PER_USER in .env (0 means no limit).

Share a chat​

You can share a conversation with a teammate, or with anyone through a link, using Share in the chat's menu in the sidebar. A shared chat is a read-only copy of the conversation as it stood when you shared it: the messages, artifacts, images, video, audio, files and links. Tool calls and thinking are left out unless you tick Include the tool calls and thinking blocks, and any token or key in them is replaced with [redacted]. Later messages don't reach the copy, so share again to send a newer version.

A teammate gets a notification and opens the copy from Shared with you in the notifications panel. Someone who opens a link sees the chat under a header that says who shared it, when, and that it's read-only. Like an app's link, a chat link has a password by default and an expiry date. A share can hold up to 50 MB, and each person can keep up to 50 shares active at a time.

You can share your own chats. On a Shared only agent, whose chats belong to the whole team, sharing requires the editor or manager role. An admin can share any chat. Task runs can't be shared.

Talk to your agents​

Prefer to speak? Dictate into any chat, hear replies read aloud, or hold a live, phone-style voice conversation — and start one hands-free with "Hey OtoDock". See Voice.

Next steps​

  • Tasks → — put an agent's work on a schedule.
  • Memory → — help agents remember across conversations.
  • Voice → — speak to your agents and hear them reply.
  • Permissions → — control what agents do on their own.