Skip to main content

The chat interface

The chat surface is where you talk to Autobot. It's a familiar back-and-forth: you type a message, Autobot replies, and the whole exchange is kept as a thread you can return to later.

The page has two main regions:

  • History panel (left) — your list of conversations, newest first.
  • Conversation + input (center/bottom) — the message stream and the box where you type.

Real-time streaming

Autobot's answers don't arrive all at once — they stream in token-by-token, so you see the reply forming as the model generates it. This matters most on longer answers and multi-step tasks: you get immediate feedback instead of staring at a spinner.

If a stream cuts off

A streamed reply can be interrupted if your network drops the connection or your sign-in session expires during a long turn. The partial message is discarded — only completed turns are saved. Just refresh and re-ask. See troubleshooting.

Inline tool-call badges

When Autobot does something beyond talking — reading your library, creating a script, building a workflow — it uses a tool, and you see that happen inline. A badge appears in the conversation, for example:

🔧 Creating script: ping_servers.py

🔧 Listing your workflows

The result of the tool appears immediately after the badge, and then Autobot continues its reply. This makes its actions transparent: you always know what it touched and when, rather than getting a finished artifact with no trace of how it got there.

A single message can trigger several tool calls in a row (for example, "create a workflow with three scripts" might list your scripts, create each one, then assemble the workflow). Each one shows its own badge. There's a built-in ceiling on how many rounds one message can take — see Limits.

Modes

Each chat runs in one of three modes, which control how much Autobot is allowed to do. You pick the mode in the chat, and it acts as a hard floor — Autobot is never offered an action above the current mode.

ModeWhat it unlocks
ResearchRead-only: list and read scripts/workflows, look up vault metadata, investigate past runs.
GenerationResearch plus creating and updating scripts and workflows.
ExecutionGeneration plus running, re-running, and investigating live runs.

Research and Generation are available to everyone. Execution mode is gated — it requires your own LLM key, and its button is hidden until you've connected one. Everything execution mode unlocks is covered on the Execution copilot page.

Pick the lowest mode that does the job

Staying in Research or Generation when you're only reading or drafting keeps you well clear of anything that touches real servers. Switch up to Execution deliberately, when you actually want to run something.

Threads, history, and archive

Every conversation is a thread. Threads keep full history, so you can pick one back up days later with all the context intact.

Active chats

The History panel lists your active threads, newest first. Each row has a menu:

ActionWhat it does
RenameChange the auto-generated title to something memorable.
ArchiveHide the thread from the active list — without losing it.
DeletePermanently remove the thread and all its messages.
Delete is permanent

Archive hides a thread; Delete destroys it. If you might want to revisit a conversation later, archive it instead of deleting.

Archived chats

Archiving doesn't delete. Archived threads stick around so you can revisit decisions or copy a turn into a new chat.

  • Open the Autobot Dashboard and click View archived chats.
  • From the archived view, each thread can be Unarchived (returns it to History) or Deleted.
  • You can still open an archived thread to read it, but you can't send new messages to it. The chat surface shows a read-only banner with an Unarchive button — click it to resume the conversation.
Where the Archived page lives

The Archived Chats page is intentionally not in the main navigation. It's reachable only from the dashboard, to keep your sidebar focused on active work.

What's next