# Settings Open Settings with the gear in the library. Settings apply to every project and are saved on your computer. Settings has five pages, listed on the left: **AI editor**, **Coding agents**, **Transcription**, **Updates** and **Troubleshooting**. Under each name is a few words on where it stands, like how many agents are connected or whether an update is ready, so you can see it without opening the page. Choose a name to open its page. Settings opens on the page you had open last. One more setting lives where you use it: the list of names and words to spell right is at the top of the transcript. See [Edit the transcript](https://clipcat.followusai.com/docs/transcript.md). It's kept for every project too. ## AI editor Pick the AI that suggests edits, shorts, the write-up and spelling fixes. Each one has a row with a button to pick it, a line that says whether it can run now, and its model. Only one is picked at a time. **On your plan**: - **Claude Code** and **Codex** run the `claude` or `codex` command on your computer, signed in to your own plan. **Model**, for Claude Code and Codex: leave it on the default to use the agent's own default model, or pick one. Codex lists the models your login can use. If one can't run, its row says why. Fix it, then choose **Check again**. Their rows also have the same install and update line as on the Coding agents page. **With your own key**: paste a key and choose **Save key**. ClipCat tries the key with one call before saving it, then keeps it encrypted with your system's keychain. **Remove key** deletes it. Each run is billed to that account. - **Claude API**: a key from console.anthropic.com. Its model is Opus 5.5 (best), Sonnet 5.5, or Haiku 5.5 (cheapest). - **OpenAI**: a key from platform.openai.com, to run GPT models. - **OpenRouter**: a key from openrouter.ai, to run any model OpenRouter serves that can use tools, from Anthropic, Google, Meta, Mistral, DeepSeek, Qwen and others. Results vary a lot by model: small models often lose track of the transcript's word numbers. A model that can't see pictures can't design slides, and the list marks those "text only". - **Model**, for OpenAI and OpenRouter: the list comes from your key once it's saved. OpenAI starts on its best model. OpenRouter waits for you to pick one. The list on the left shows which AI is picked, and says "not set up" when it can't run yet. The **Edit** tab of a project can switch between them too. See [Review the suggested edit](https://clipcat.followusai.com/docs/ai-editor.md). ## Coding agents Connect Claude Code, Codex, the Claude desktop app, Cursor, VS Code, Windsurf, LM Studio or Gemini CLI, and it can edit your projects while ClipCat is open. Under **Connections**, agents on this computer are listed first, and the rest are folded under **Not found on this computer**. Each agent has a row: - The line under its name says whether it's connected, when it last connected, or that it isn't on this computer. You can connect an agent before you install it. - The path is the agent's own settings file, where ClipCat adds its entry. - **Connect** shows the exact entry ClipCat will add and asks first. **Add to** (the agent's name) adds it, then ClipCat checks the connection. - **Test** checks the connection again, step by step: the file is read back, the entry is run through to ClipCat, and the agent's own command is asked. The last step turns to done once the agent really connects. - **Disconnect** takes ClipCat's entry out of the file. Nothing else in the file changes. **Set up another MCP app**, under the list, shows the entry for any other app that runs local MCP servers: the JSON, and the command, arguments and environment, each with a copy button. See [Use ClipCat from your coding agent](https://clipcat.followusai.com/docs/agents.md#set-up-another-mcp-app). Under **How to use it**, **Steps for** picks the agent to show the steps for. ### Install or update Claude Code, Codex and Gemini CLI The Claude Code, Codex and Gemini CLI rows also show the version of the command-line tool on your computer next to the newest release. - **Install Claude Code** or **Install Codex** runs its maker's own install script. **Install Gemini CLI** runs `npm install -g @google/gemini-cli`, or `brew install gemini-cli` if you have Homebrew but not npm. With neither, the row says to install Node.js first. - **Update to** (a version) updates it the way it was installed: its own updater, npm, bun, Homebrew or winget, so you don't end up with two copies. - Next to the button is the exact command it runs. Its output shows while it runs. - A copy that came inside another app is left to that app, and the row says how to update it. - The desktop apps (the Claude desktop app, Cursor, VS Code, Windsurf and LM Studio) can't be installed from ClipCat. When one isn't on your computer, its row has **Get** (the app's name), which opens its maker's download page. - If your Codex is older than another Codex on your computer (usually the one inside the ChatGPT app), the row warns that the newer one's settings may not work in it. The **How to use it** part of the page, under the agents, walks through the first session for each agent, with prompts to copy. See [Use ClipCat from your coding agent](https://clipcat.followusai.com/docs/agents.md). ## Transcription - **Language for new projects**: the language ClipCat transcribes new projects in. English unless you change it. **Detect automatically** has whisper guess the language from the first seconds of speech, and skips Parakeet. Each project can change its own language from the transcript. - **Whisper model**: **Large v3 Turbo** (most accurate, the default) or **Base** (fastest, less accurate). - **Engine** (Apple Silicon only): **Parakeet when it knows the language** (the default) or **Always whisper**. The choices above are under **New projects**. Under **Speech models on this computer**, each speech model has a row that says whether it's downloaded and how big it is. - **Download** gets it now, with progress, so the first transcription doesn't wait. Otherwise it downloads the first time it's needed. - **Delete** removes it to free the space. It downloads again when needed. A model is loaded only while it transcribes. See [Add videos and transcribe them](https://clipcat.followusai.com/docs/importing.md). ## Updates ClipCat checks for a new version when it opens and every four hours after. The Updates page says what it found. **Check for Updates** in the ClipCat menu (in the Help menu on Windows) checks now and opens this page. - **Check now** checks again. - **Restart to update** shows when a new version has downloaded. The status line at the foot of the window shows it too. - **Open the release page** shows when this copy can't install the update itself. Download the new version there. A copy you built yourself doesn't update. ## Troubleshooting - **Save diagnostics** saves one zip: ClipCat's log, its versions and setup, recent jobs, each project's history and crash reports. It holds no video, sound or transcript, and your home folder's name is replaced with `~`. Attach it to a bug report. - **Show the log** opens the folder with ClipCat's log. See [Fix a problem](https://clipcat.followusai.com/docs/troubleshooting.md). Last updated 29 October 2026. Web page: https://clipcat.followusai.com/docs/settings