Transcript, thinking & tools
The transcript is where you read what your agents are doing. It’s built with native views, not a web page, so it stays fast even in long chats.
Streaming replies
Section titled “Streaming replies”Replies stream in as the agent writes them. Markdown is fully supported, including tables and code blocks with a Copy button.
Every message ends with a Copy button and its details: the model that wrote it and the full date and time it was sent.
Thinking
Section titled “Thinking”OpenClaw only stores the agent’s reasoning when the session’s reasoning level is on. You can set it from:
- the hint above the composer;
- the chat’s ⋯ menu: Gateway Reasoning → Save & Stream;
- the chat itself, by sending
/reasoning on.
| Gateway Reasoning | Effect |
|---|---|
| Save & Stream | Reasoning streams live and is saved in the history. |
| Stream Only | Reasoning streams live but isn’t saved. |
| Off | No reasoning. This also hides reasoning text in the transcript. |
Thinking steps
Section titled “Thinking steps”How much of the agent’s thinking steps (reasoning and tool calls) you see is a separate choice, set per device under Thinking Steps in the chat’s ⋯ menu or in Settings:
- None: only replies.
- Live Only (default): thinking and tool calls show while the agent works, then hide once the reply is done.
- All: every turn keeps its thinking and tool calls, folded into one collapsible Thinking item once the turn finishes.
Copy Thinking stays in a reply’s context menu whichever you pick.
Tool cards
Section titled “Tool cards”Each tool call is shown as a card with its arguments and result. When a tool call starts a helper (subagent) run, the card has an Open run button that takes you to it.
Images
Section titled “Images”Pincer shows images inline, with a Quick Look-style preview and sharing. This covers:
- attachments on messages;
- the agent’s
MEDIA:lines, the same as the web UI. Local files are fetched through the gateway’sassistant-mediaroute; - public
httpsimages the agent links to. These are downloaded directly, without any credentials or cookies.
To stop Pincer from loading images from the web, turn off Load images the agent links from the web in Settings.
Your messages
Section titled “Your messages”Messages from you appear under your own name, even when you sent them through Discord. A small via Discord tag shows where they came from. Set your name in Settings. It defaults to your macOS full name.
Models
Section titled “Models”The chat toolbar shows the session’s model. Pick another from the list the gateway offers, or go back to the agent’s default, and new messages use it.
Each reply’s footer shows the model the gateway recorded for it, so earlier replies keep their original model after you switch.
History and caching
Section titled “History and caching”The full history of every chat loads in the background, so scrolling up never waits for the network. After connecting, Pincer caches every chat, most recently active first, and skips chats that haven’t changed. Opening a chat shows the cached transcript right away, then fetches only what’s new.
The cache location and how to turn it off are described in Security & privacy.