Website: thuki.app
Commands are written as whole-word / triggers anywhere in your message. Press / to open the command suggestion menu, then Tab to complete or Enter to select.
Commands can be combined when their behavior allows it. For example, /screen /think captures the screen and enables extended reasoning, while /think /tldr summarizes with reasoning enabled.
Transform commands (/rewrite, /refine, /translate, /tldr, /bullets, /todos, /extract) never trigger web search or the auto-search classifier. They run only on the text you supply (or OCR text). Plain messages, /explain, and /think still follow Auto search in Settings; /search always forces a live look-up. Auto search is [behavior].auto_search in configurations.md; privacy and network notes in privacy.md and search-privacy.md; design handbook in built-in-web-search.md.
Commands that operate on text follow a consistent input priority:
- Highlighted text + no typed text: highlighted text is the input
- No highlighted text + typed text after command: typed text is the input
- Both present: highlighted text is the primary input; typed text is appended as an additional instruction
This means you can highlight text anywhere on screen, summon Thuki with double-tap Control, type a command, and hit Enter without retyping the selected content.
/extract, /tldr, /translate, /rewrite, /refine, /bullets, /todos, and /explain pull the text out of attached images locally via macOS Vision OCR, so they work even when the active model has no vision capability. They read text, not scenes: describing a textless image still needs a vision model. Only plain submits and /screen alone send the image itself to the model. See OCR-supported commands for the full list and details.
Forces a live web look-up and answers with citations (zero-setup).
Usage: /search <question>
Examples:
/search who owns Figma now?: searches live sources for a current answer/search latest React 19 release notes: retrieves recent release information from the web
Behavior: Forces engines-only web search even when Auto search is off or would skip. Answers are grounded in retrieved sources with inline citations and a Sources footer. With Auto search on (Settings → Behavior, default), plain messages may also search when live facts are needed.
Limit: Use when you want a forced look-up, Auto search is off, or a plain turn skipped the web.
Extracts all visible text from screenshots or attached images using macOS Vision OCR.
Usage: /extract [optional message]
Examples:
/extractwith an attached image: extracts all text from the image/screen /extract: captures the screen and extracts all visible text
Behavior: Text is extracted using the macOS Vision framework and returned verbatim in a code block. No prose or explanation is added. When multiple images are provided, each result is separated by a horizontal rule. Returns "[No text detected]" when no readable text is found. Does not trigger web search.
Composable: /extract can combine with /screen to capture then extract in one step.
Permission: Uses the same Screen Recording permission as /screen when combined with it.
Captures your screen and attaches it as context for the current message.
Usage: /screen [optional message]
Examples:
/screen: sends a screenshot with no additional message/screen what is this error?: attaches a screenshot and asks a question about it
Behavior: The screenshot is taken when you submit the message. Thuki's own window is excluded from the capture, and the image appears in your message bubble like a pasted screenshot.
Composable: /screen can combine with /think and utility commands. For example, /screen /rewrite captures the screen and rewrites whatever text the model can see.
Limit: One /screen capture per message. You may also attach up to 3 images manually for a total of 4 images per message.
Permission: Requires Screen Recording permission. If denied, Thuki cannot capture the screen until access is granted in System Settings.
Enables extended reasoning before the model responds.
Usage: /think [optional message or highlighted text]
Examples:
/thinkwith highlighted text: reasons through the selected content/think what are the tradeoffs of a monorepo vs polyrepo?: asks a question with deep reasoning enabled
Behavior: A collapsible Reasoning block appears above the response showing the model's reasoning chain. The final answer appears below it as normal.
Composable: /think works with /screen and all utility commands. For example, /think /tldr summarizes with extended reasoning enabled.
Translates text to another language.
Usage: /translate [language] [text] or /translate with highlighted text
Examples:
/translatewith highlighted text: auto-detects the source language and translates it/translate jawith highlighted text: translates highlighted text to Japanese/translate Spanish meeting notes here: translates typed text to Spanish
Behavior: Outputs only the translation with no commentary or explanation. Does not trigger web search.
Composable: /translate works with attached images or /screen. Vision OCR extracts the text first; translation runs on the result. Omitting a target language defaults to Vietnamese.
Language format: The target language can be a full name (French), ISO code (fr, fra), or common shorthand.
Default behavior: If no language is specified, the text is translated to Vietnamese.
Rewrites text to sound like natural, casual, everyday speech.
Usage: /rewrite [text] or /rewrite with highlighted text
Examples:
/rewritewith highlighted text: rewrites the selected text/rewrite so basically what happened was i was trying to fix the bug: rewrites typed text in a natural, casual voice
Behavior: Rewrites text to sound like a fluent native speaker talking day to day: relaxed and casual by default, while keeping your meaning, personality, and point of view. It mirrors your original formatting instead of flattening it: Markdown headings, bold, lists, links, blockquotes, and code all stay, and only the text inside them is improved. Quoted lines, code, URLs, @mentions, #channels, and the emoji or expressive spellings you used are kept exactly as written. It fixes what reads awkwardly and leaves what already reads well alone, only stays formal when the original clearly is, and will not bolt a "we" voice onto an imperative or impersonal note. Outputs only the rewritten text. A Replace button on the result writes the rewritten text straight back into the app you were using, replacing your selection; turn on auto-replace in Settings to skip the button. Follow-up tweaks in the same chat, like asking for a longer or more formal version, keep the Replace button too. Does not trigger web search.
Composable: /rewrite works with attached images or /screen. Vision OCR extracts the text first, then rewrites it.
Summarizes text into 1-3 short, direct sentences.
Usage: /tldr [text] or /tldr with highlighted text
Examples:
/tldrwith highlighted text: summarizes the selected content/tldr [paste a long article]: summarizes typed or pasted text
Behavior: Captures the core message, key decision, or critical takeaway. Skips background detail and qualifications. Does not trigger web search.
Composable: /tldr works with attached images or /screen. Vision OCR extracts the text first, then summarizes it.
Fixes grammar, spelling, and punctuation while preserving your voice.
Usage: /refine [text] or /refine with highlighted text
Examples:
/refinewith highlighted text: corrects the selected text/refine hey just wanted to follow up on the thing we discussed: cleans up typed text
Behavior: Corrects errors and smooths rough phrasing without restructuring or adding new ideas. Your original tone and meaning stay intact. A Replace button on the result writes the refined text straight back into the app you were using, replacing your selection; turn on auto-replace in Settings to skip the button. Follow-up tweaks in the same chat, like asking for a longer or more formal version, keep the Replace button too. Does not trigger web search.
Composable: /refine works with attached images or /screen. Vision OCR extracts the text first, then refines it.
Extracts key points from text as a markdown bullet list.
Usage: /bullets [text] or /bullets with highlighted text
Examples:
/bulletswith highlighted text: extracts key points from the selection/bullets [paste meeting notes]: extracts key points from typed or pasted content
Behavior: Each point is a concise, self-contained statement. Ordered by importance or logical sequence. Filler and repetition are removed. Output uses - prefixed markdown bullets. Does not trigger web search.
Composable: /bullets works with attached images or /screen. Vision OCR extracts the text first, then extracts key points.
Explains any concept, term, or code snippet in plain language, always with a concrete example.
Usage: /explain [text] or /explain with highlighted text
Examples:
/explainwith highlighted code: explains what the code does and why/explain what is a closure?: explains the concept with a concrete example/explain JWT: breaks down the term with a real-world analogy and example
Behavior: Outputs a brief explanation followed by at least one concrete example. Assumes no background knowledge. Skips jargon or defines it when unavoidable. No intro or sign-off.
Composable: /explain works with attached images or /screen. Vision OCR extracts the text first, then explains it.
Summarizes what a piece of text is about, then extracts every task, action item, and commitment as a markdown checkbox list.
Usage: /todos [text] or /todos with highlighted text
Examples:
/todoswith highlighted text: summarizes and extracts to-dos from the selected text/todos [paste a conversation or notes]: processes typed or pasted content
Behavior: Responds in two parts: a short paragraph explaining the context and what is at stake, followed by a - [ ] checkbox list of all tasks. Each to-do includes who is responsible, plus any deadline or timeframe if mentioned. Does not trigger web search.
Composable: /todos works with attached images or /screen. Vision OCR extracts the text first, then extracts to-dos.