Icons

Prompt input

The composer a person types a prompt into

Attachment

A file clipped to a prompt or message

Model picker

Switching models mid-conversation

Prompt enhance

One tap that rewrites a rough draft into a better prompt

Slash command

A typed shortcut to canned prompts and actions

Prompt template

A saved prompt with {{variables}} to fill in

Message

A single turn in a conversation

Streaming

A reply arriving token by token, caret trailing the text

User turn

The person's side of the conversation

Rendered reply

Markdown output — headings, lists, code blocks

Message actions

Copy, share and feedback under a reply

Regenerate

Ask the model for another answer to the same prompt

Model error

The model failed, refused, or was rate limited

Stop generation

Halting a reply while it is still streaming

Edit message

Changing a sent prompt and regenerating from it

Response versions

The 2/3 pager between regenerations of one answer

Share conversation

A link to this conversation for someone else

Chat history

Past conversations, grouped and searchable

Pinned conversation

Keeping a conversation at the top of the list

New chat

Starting a fresh conversation

Artifact

Generated output substantial enough to get its own panel

Reasoning

The model's visible chain of thought

Thinking

Working, before the first token arrives

Reasoning steps

A trace broken into ordered, inspectable steps

Reasoning effort

How much the model should think before answering

Tool call

The model invoking a tool

Web search

A search of the web, run as a tool call

Code execution

Code run in a sandbox, with its output

Tool toggle

Turning an individual tool on or off

Generative UI

A component rendered from a tool result, not text

Media generation

An image, audio or video the model produced

Source

A document the answer was grounded in

Citation

An inline marker pointing at a source

Citation popover

The passage behind a citation marker, on hover

Agent

An autonomous run working on its own

Approval gate

A step that waits for a human to allow or deny it

Agent plan

The checklist an agent writes for itself

Agent handoff

One agent passing the task to another

Agent timeline

The activity log of a long agent run

Multi-agent

Several agents working in parallel

Interrupt agent

Cutting into an agent mid-run to redirect it

Scheduled agent

An agent that runs on a timetable, not a prompt

Agent connector

An external system the agent can act in, wired up

Response rating

Thumbs up or down on a reply, with a reason

Inline correction

Fixing the answer where it is wrong

Output comparison

Two answers side by side, labels hidden until picked

Eval results

Pass rates across a test suite

Edit diff

An AI-proposed edit, accepted or rejected per hunk

File upload

Getting a file into the conversation

Files in context

What the model can currently read

Context window

How much of the model's context budget is used

Context picker

The @-mention menu for adding context

Voice input

Speaking to the model instead of typing

Waveform

Live audio level while listening or speaking

Voice conversation

A spoken exchange with turns and replay

Voice call

A live call with the assistant — mute, timer, end

Read aloud

The reply spoken as audio

Barge-in

Talking over the assistant to cut it off

Memory

Facts the assistant keeps about you

Memory suggestion

The assistant proposing to remember something

Saved to memory

Confirmation that a fact was kept

Memory toast

The receipt shown when a fact is saved

Model

The model itself, and choosing between them

Token

The unit a model reads, writes and bills in

Cost

What this reply or this conversation cost

Latency

How long until the first token arrives

Sampling

How much randomness the model answers with

Refusal

The assistant declining a request, with the redirect still on the table

Moderation flag

The content filter interrupting a prompt or a stream

Confidence

How sure the answer is, flagged where the claim is made

PII redaction

Personal data stripped before the model sees it

Prompt injection

Instructions smuggled in through retrieved content

Hallucination warning

A flag on claims the model may have invented