Answer, Assist, Recap, and screen actions
What each action sends to Observer and when to use it.
Every action in Observer packages a different set of inputs before the AI responds. Picking the right one keeps answers focused and fast.
Shared context on every action
Observer chat always carries your session context forward. Alongside each action's primary inputs, Observer also sends:
- The active Instruction Mode prompt (when a preset is selected)
- The PDF summary from attached instruction files
- Your chat history from the current meeting
This is how Observer keeps continuity across Answer, Assist, Recap, and typed messages throughout the call.
Answer (⌘ D)
What Observer sends: The live meeting transcript is the primary input. Instruction Mode prompt, PDF summary, and chat history are included as well.
What happens internally: Observer sends recent transcript turns with an Answer prompt. The model reads what was just said and returns a short reply you can use immediately.
Best for: Direct questions from the conversation — not what is on your screen.
Requires: Transcript running (microphone permission and an active meeting session).
Assist (⌘ ⇧ D)
What Observer sends: A fresh screenshot is primary, plus the transcript when listening is active. Instruction Mode prompt, PDF summary, and chat history are included as well.
What happens internally: Observer captures your screen, attaches transcript context if available, and sends both with an Assist prompt. The screen is treated as primary evidence; the transcript supports what was asked.
Best for: Questions about slides, code, diagrams, dashboards, or anything visible on screen.
Without transcript: Assist still works screen-only if listening has not started yet.
Recap (⌘ R)
What Observer sends: The live transcript is primary. Instruction Mode prompt, PDF summary, and chat history are included as well.
What happens internally: Observer asks the model to summarize recent conversation turns — useful when you joined late or lost track of the thread.
Best for: Catching up on what was discussed in the last few minutes.
Requires: Transcript running.
Screen analysis (⌘ ↵)
What Observer sends: A screenshot only — no typed question. Instruction Mode prompt, PDF summary, and chat history are included as well.
What happens internally: Same as pressing Enter on an empty chat input. Observer captures the screen and asks the model to explain or solve what is visible.
Best for: Quick help with code, errors, or slide content without typing a prompt.
Ask in chat (type + Enter)
What Observer sends: Your typed message with whatever context the current chat mode includes. Instruction Mode prompt, PDF summary, and chat history are always included.
What happens internally: Default chat uses meeting context already in the session. The exact inputs depend on whether Screen + text or Power Mode is toggled on.
Best for: Custom questions when you want full control over wording.
Screen + text ( in chat)
What Observer sends: Your typed question plus a fresh screenshot on every send. Instruction Mode prompt, PDF summary, and chat history are included as well.
What happens internally: Toggle the icon in the chat bar. While it is active, each message includes screen capture + your prompt. This is not the same as Power Mode — it does not include the live transcript unless you also enable Power Mode.
Best for: Specific on-screen questions where you want to type exactly what to analyze.
Power Mode ( in chat)
What Observer sends: Transcript, screenshot, Instruction Mode prompt, PDF summary, your typed question, and chat history together.
What happens internally: Toggle the icon when a transcript is available. Observer merges every major context source into one request so the model can reason across conversation, visuals, and your preset instructions.
Best for: Complex questions that need conversation history, screen content, and uploaded reference material at once.
Requires: Transcript running (the icon is disabled until listening is active).
Quick decision guide
- Conversation-only → Answer or Recap
- Screen matters → Assist or Screen + text
- Everything at once → Power Mode
- No prompt, just analyze what's visible → Screen analysis (⌘ ↵)