How to use CSuite.

CSuite is a native desktop app for macOS, Windows and Linux. Everything runs on your computer. There’s no account to create and no website to log into, and the app speaks your language: it ships in 20 languages and follows your operating system’s language on first launch.

CSuite has a single window with four main areas:

• The menu: the strip of icons down the side of the window (the left in most languages; it mirrors to the right in Arabic and Hebrew). The CSuite mark at the top opens the Dashboard; below it sit the four modalities (Text, Image, Audio, Video), then Compositions, Chat and Workflows, and finally Models and Account. While an update is downloading or ready to install, a small update button appears between Workflows and Models. Hover an icon to see its label.
• The Dashboard: a Start creating row (Text, Image, Audio, Video, Chat, Workflow), a Quick actions row (change the project folder, download models, add provider keys, manage models, open Account) and a Files section grouping everything you’ve made: text, images, audio, video clips, compositions, workflows and chats, up to 20 per group with a “+N more” link. Clicking any card opens it in the workspace that owns it.
• The sidebar: the list of files in your current modality. For example, in the Image workspace it shows every image in your project folder with a small thumbnail and filename.
• The main panel: where you preview, edit and generate content. A model pill sits beside the panel’s main bar and the prompt composer sits below the preview, so you can iterate without leaving the workspace.

Each workspace is built around the same idea: pick a model from the pill (or leave it on Auto), type a prompt, and the result lands in your project folder and shows up in the sidebar. On macOS and Windows the title bar also shows a breadcrumb of where you are. ⌘+ / ⌘- / ⌘0 (Ctrl on Windows and Linux) zoom the whole interface in, out and back to normal; on macOS the View menu lists them alongside full screen.

The project folder is the heart of CSuite. Everything you generate or edit is saved there as a real file you can open in Finder, File Explorer, or any other app.

• The app starts with a default folder inside your Documents, so there’s nothing to configure up front. A folder pill above the sidebar shows which folder you’re in; click it to switch.
• With no folder set, or one that has gone missing or become unreadable, that pill turns into an amber Select folder pill, and nothing can be generated until you answer it.
• Each workspace shows the files it can open: Text lists plain text, Markdown and rich-text (HTML) documents; Image lists PNG, JPG, WebP, GIF and SVG; Audio lists MP3, WAV, FLAC, OGG, AAC and M4A; Video lists MP4, WebM, MOV and AVI.
• Drag files from your desktop straight onto the sidebar list and they’re copied into the folder and opened; a file the workspace can’t open is refused with a message. You can also drop files into the folder from outside the app and click the refresh button next to the folder pill to pick them up.
• Once a folder holds more than 20 files, a filter box appears above the list, with a sort control (newest, oldest, or by name) tucked inside it.
• Switching to a different folder gives you a clean workspace, great for keeping client projects separate.
• Use the ⋯ menu on any file in the sidebar to reveal it in your operating system’s file manager, duplicate it, or delete it. Click the filename in the image, audio or video preview header to rename the file in place.
• Select a file and press Delete or Backspace to remove it. The app shows a confirm dialog first so you can’t lose work by accident.

An AI model is the engine that does the actual generating. CSuite lets you reach models in three ways: through a cloud platform, directly from the model’s vendor, or on your own computer.

Cloud providers

CSuite connects to three cloud platforms (Runware, OpenRouter and Replicate) that run powerful models on their own servers. Through them you reach models from OpenAI, Anthropic, Google, Black Forest Labs, ByteDance, DeepSeek, xAI and more, with a single key per platform. OpenRouter is one key and one API in front of models from OpenAI, Anthropic, Google, DeepSeek, xAI and the open-weight vendors, and it reports the exact billed cost of every generation back to the app. CSuite talks to all three directly using your own API key. Your prompts never pass through us.

Model vendors

You can also skip the platform and use a model lab’s own API with that lab’s own key. CSuite supports twelve vendors directly: OpenAI, Anthropic, Google, ElevenLabs, xAI, DeepSeek, MiniMax, Black Forest Labs, Runway, Kling, ByteDance and Fish Audio. Requests go straight from your machine to the vendor, and the vendor bills the usage to your own account with them.

Add the key under Models → Vendors the same way you would a platform key. That vendor’s models then show up in the regular model pickers next to everything else. One thing to know: Auto only routes through the three platforms and the local runtimes, so a model that exists only on a vendor key is one you pick by hand.

Local providers

Local providers run open-weight models directly on your computer. Once a model is downloaded you can use it offline, with no per-request cost, and nothing leaves your machine. CSuite supports three runtimes:

• Ollama: the easiest way to run text models like Gemma 4, Llama 3.1, Mistral Small and Qwen3.8, and the home of the local Decision models (Tev1, Nimble, Clef, D1, d3 and Decider, plus Laya and Strands Decider on an Apple silicon Mac) that read your prompts for Auto, and of Phonon, an English text-to-speech model for Apple silicon Macs. CSuite bundles Ollama for you, or uses an Ollama you already run elsewhere: open Models → Providers → Ollama, click the settings gear in its header, enter the Server URL and press Save; the app checks the connection and reports “Connected” with the number of models it found (your own server needs Ollama 0.35.1 or newer for the Decision models, and 0.40.9 for d3 and Phonon). On Apple silicon Macs Ollama also generates images: the app runs a second, pinned Ollama server just for that, sharing the same model store, and installs it the first time you pull an image model. It only works with the built-in Ollama (releases after v0.32.5 can’t run image models), so with your own server set, press Reset under Server URL to use them. A Check for updates button compares your installed models against the registry and pulls only the layers that changed.
• HuggingFace: runs ONNX-exported open models fully offline, including local text-to-speech with Supertonic TTS (ten voices) and on-device transcription with Whisper Base, Whisper Small and Whisper Large V3 Turbo. CSuite installs the runtime for you on first use. On Windows it needs the Microsoft VC++ 2015–2022 Redistributable.
• GGML: generates images and video locally, built on stable-diffusion.cpp, with no API key. CSuite downloads the runtime and model weights on first use; weight downloads can be paused, resumed, cancelled or discarded, and a freshly downloaded model shows up in the picker without a restart. It’s the most demanding of the local runtimes, so each model lists the machine it needs, a download is refused when your disk is nearly full, and there is no build at all for Intel Macs, Linux arm64 or Windows on ARM.

How well a local model performs depends on your computer’s memory and (on Macs) your Apple silicon chip. For Ollama models, CSuite compares each model’s needs against your machine and blocks the download if it needs more RAM than you have; Ollama’s image models, Laya and Strands Decider need a Mac with Apple silicon, so on Intel Macs, Windows and Linux their download is blocked with an explanation. Interrupted downloads pick up where they left off after a restart. On Linux, the bundled Ollama runtime needs zstd installed before it can unpack itself. The settings gear of each local provider lists its pieces under Runtime with the version CSuite uses: a missing one offers Download, and when an app update moves to a newer release than the one installed, Update to … installs it now instead of the next time the runtime starts.

Neocloud

Neocloud providers are GPU clouds you rent by the second to run the local models above on a GPU instead of your computer, starting with RunPod. Open Models → Providers → RunPod, paste an API key from your RunPod account and click Save key; the app checks it right away. The Text, Image, Audio and Video workspaces then list every local model that runs on a Linux GPU under RunPod in their pickers. Chat, workflows and compositions don’t offer it.

• Renting. A run that needs a new GPU asks first, showing the card and its hourly price (tick Don’t ask again to skip the question). CSuite offers RunPod the cheapest cards with enough memory for the model on its Secure Cloud; when the cheapest is taken, RunPod places a similar one, so the question shows the highest rate you could pay too. RunPod bills your account by the second, and Analytics shows the spend under RunPod, failed and stopped runs included.
• Kept ready. After a job the GPU stays for 10 minutes (5, 10, 15 or 30 under Keep the GPU ready on the RunPod page), so a follow-up, such as editing the image you just made, starts at once and reuses the models already on it. Runs take turns on the GPU: one that waits shows “Waiting for the GPU…”. A model that needs more memory swaps it for a bigger card. The time left shows in the title bar on macOS and Windows (click it to open the RunPod page) and on the RunPod page everywhere; Release now removes the GPU at once, and quitting CSuite removes it too.
• Setup and downloads. CSuite sets up each new GPU and downloads the model to it again; nothing is kept between GPUs. Setup takes about a minute; if the ready-made image and video runtime can’t be used on a GPU, CSuite builds one there instead, which adds 5 to 20 minutes to its first image or video. Your prompts and files reach the GPU through an encrypted SSH tunnel and stay on RunPod’s machine only while it runs.

Click the Models icon in the menu. It has three tabs: Defaults, Providers and Vendors.

Defaults

The first tab, and the one you’ll come back to. It lists one row per thing you can make, with the model each currently uses, in four groups: Writing (Text, Chat), Visual (Image, Video), Audio (Text to speech, Music, Sound effects, Transcribe) and Decision. Every row starts on Auto. Pick a row to change its model and tune that model’s settings; Chat and Decision are a model pick only, with nothing to tune.

This tab is the only place that writes the global pick. Choosing a model from the pill inside a workspace is a session override: it holds until you close the app, switch modality or change the default itself, then the default takes over again.

Decision

Decision models don’t write a reply. They answer a typed question about a piece of text (a choice, a yes or no, a score) with a probability for each option. CSuite uses yours to read what a prompt asks for, in any language, so Auto can pick the right model, so Chat can choose the right tool, and so the Text workspace knows a request from an edit. It also chooses between options in workflow Decide nodes. The catalog has ones on Ollama (Tev1, Nimble, Clef, D1, D1 Omni, d3 and Decider, and Laya and Strands Decider on an Apple silicon Mac), plus Jev, Drex and others on Runware and OpenRouter. GPT 6 Luna Decisions (on OpenRouter or your own OpenAI key), Decider V1.2, D1, D1 Ultra, Span-02 and Clef Omni read images as well as text, and so does d3 on your own machine, so a Decide node can choose between pictures with them. Under Auto an installed local decision model always wins over a cloud one, so your prompts stay on your machine, and when no Decision model is available your Text model takes over the job.

Providers

The second tab is where keys and downloads live, in three groups: Cloud (one entry per cloud platform), Local (Ollama, HuggingFace and GGML) and Neocloud (RunPod). Click a group’s name to fold it away; the app remembers which ones you folded. The info icon at the end of each group’s name says what that kind of provider is. From here you can:

• Add, reveal, replace or remove a cloud provider’s API key.
• Browse each provider’s model cards, newest first within each modality, with a search box to filter them by name. Cloud cards show Price, Age (how long ago the model was released, with the month) and Info; local cards swap the price for Size and Req. (the machine the model needs), and Ollama cards add Context (how much text the model can hold), so you can tell what a model costs and whether your computer can run it before you enable it.
• See what a model makes before you use or download it: its card has a Samples button wherever samples exist, on cloud and local cards here and on Vendors. It opens real outputs from that model (images, clips, audio, transcripts, written answers or Decision answers), made from the same prompts sent to every model of its type, so two models compare like for like. Transcription models are all given the same three clips instead. The first line says where the samples ran: a local model’s usually ran on its own runtime (Ollama, GGML or Hugging Face), and a few that are also hosted show the hosted run. You don’t need a key or a download to look. Images, video and audio load from the internet, so if some don’t appear, check your connection and open Samples again.
• Download a local model: click Download and watch the progress bar. You can keep using the rest of the app while it downloads, and pull several Ollama models at once (GGML downloads run one at a time).
• See which models are already installed and how much disk space they’re using.
• Hide models you don’t want cluttering up the picker when you’re generating content.
• Delete models you no longer need to free up disk space.
• A model offered by more than one platform lives under one at a time: enabling it under a second provider asks Switch model provider? and moves it across.

Vendors

The third tab lists the model labs you can use directly with their own key, such as OpenAI, Anthropic, Google and more. Add a key here and every call to that vendor’s models runs through its official API. Their models join the regular list, and an Add from button takes more: search the vendor’s own model list where it publishes one (OpenAI, Anthropic, Google, ElevenLabs, xAI, DeepSeek, ByteDance), or paste a model id where it doesn’t.

The model picker

Every model pill in the app opens the same picker: a search box, a dropdown that shows All, Cloud or Local models, a clock button that lists your last five picks (Recent models), and a Manage models button that jumps straight back to this screen.

Each row names the model and its provider, and small icons at the end of the row show what the model supports, such as Layout boxes, Reference images, a Start frame or Tool calling. Rest the pointer on the icons to read their names, or type a feature such as “lyrics” into the search box to list only the models that have it.

Auto and prompt settings

Auto is the first row in every model picker (marked ✦) and the default for every slot on a fresh install. Instead of you choosing a model, it reads the prompt and picks one: a logo prompt goes to a vector model, for example. The reading is done by your Decision model, so the prompt can be in any language. Anything you attach is a hard requirement, so a start frame or a reference image only ever lands on a model that takes one. When several platforms are connected it prefers Runware, then OpenRouter, then Replicate; it also uses your installed Ollama and HuggingFace models, but never a vendor-only model, a GGML-only model or a model you added yourself.

• Under Auto the per-model settings panel is hidden and the cost estimate next to Generate is suppressed, because neither is known until the model is chosen.
• Settings can ride along in the prompt instead: “red car, 16:9” sets the aspect ratio, “8 seconds, with audio” sets a video’s duration and sound. The fragment is stripped from what the model sees, applied for that run only, and checked against what the chosen model actually supports. These shorthands are English-only and work while the slot is on Auto.
• If the best pick needs a platform you haven’t connected, a dialog titled Auto found a better model offers Continue with the best model you can run now, or Add a provider instead. Your answer is remembered for the session, so it won’t nag.
• If a platform turns a request down because its key was rejected, it’s out of credit or it’s rate limited, Auto moves that generation to the next best model on another platform and says so in a Switched model note. It skips that platform for ten minutes (one for a rate limit), or until you save or re-test its key. A model you picked yourself is never swapped.

Add any model

The curated catalog isn’t a ceiling. On Models → Providers, each cloud platform’s list has an Add from Runware, Add from OpenRouter or Add from Replicate button in its header. Search the platform’s full library (or paste a model URL or its owner/name), preview it, optionally set the price and add it. Your Chat model reads the listing and fills in the rest: type, label, company, release month, whether it supports tools and where Auto should rank it. Added models carry an Added by you badge, sort first, have Refresh and Remove actions, and retire themselves when the built-in catalog ships the same model. A built-in model the catalog has since hidden shows a Hidden badge in the same search and can be added back. Leave the price blank and the model’s runs are simply left out of cost estimates and spend totals. Decision models are curated only and can’t be added this way.

More settings

The model popover in every workspace opens from the pill that shows the current model’s name, and holds the controls the catalog gives a first-class home: aspect ratio, duration, voice, and so on. Everything else a model accepts (seed, steps, LoRA strength, frame rate, output quality, watermark…) sits under a More settings accordion in the same popover, and per-run file and text inputs appear as chips beside the prompt.

The text workspace is a rich-text editor for drafting, rewriting and brainstorming with AI assistance. Create new document offers three kinds: Rich text (saved as HTML), Markdown and Plain text, and the sidebar opens any existing file of those types. Every document is an ordinary file in your project folder, saved automatically half a second after each change.

Writing with AI

The prompt box also holds the wand, which sharpens a rough instruction before you send it, and Add context, which attaches up to five reference files or web pages the model reads as source material. Your last six instructions on a document stay as reusable chips above the box, and when you pause typing the editor can suggest the next line in faint grey (press Tab to accept).

A real editor underneath

• Format ribbon, slash menu (/), find (⌘F), an Outline of your headings and live word, character and token counts.
• Pages: split a long document into pages, reorder them, or ask for “a 5 part guide” and CSuite writes it page by page.
• Generate image from a selection, and drop images, audio or video straight into a Markdown or rich-text document.
• Export as Markdown, HTML, plain text, PDF or EPUB; multi-page documents get a contents page, chapters or an index file.

coastline-dawn.png

RunwareFlux 2 Pro16:9PNG+ reference

cinematic coastline at golden hour, soft morning mist…Generate

The image workspace is built around a thumbnail grid sidebar and a large preview panel with an editing toolbar across its top. A zoom pill in the preview goes from 10% to 1600%, with 1:1 and Fit buttons, and ⌘ / Ctrl + scroll (or a pinch) zooms too. Every image is a plain file in your project folder; drag your own in and they work like generated ones.

Generating an image

Editing and AI tools

• AI menu: Upscale 2×, Upscale 4× and Remove background on the open image, run through Runware or Replicate, saved as a new file.
• Resize, Crop, Adjust, Rotate, Annotate, Background (a gradient, colour or picture behind the image, with padding, rounded corners and a shadow), Quality and Format: a built-in canvas editor that runs on your machine and saves in place or as a new copy.
• Info shows dimensions, format and, for images CSuite made, the model and prompt, with Use prompt & settings to iterate on an old result. Copy image on a row puts it on the clipboard.
• P-Image Upscale is a model rather than a menu item: pick it and the prompt box steps aside, the button reads Upscale, and you attach the image to enlarge.
• Layout appears on Flux 3 Image, Flux 3 Image Klein and MOSS-Image: draw a box for each element on a canvas, or press Suggest to draft them from your prompt. On Flux 3, boxes over a selected image keep, move or remove what is there.

The audio workspace handles voice, music and sound effects, and transcribes recordings into text, from cloud models or entirely offline. It opens MP3, WAV, FLAC, OGG, AAC and M4A files, and a Type control in the Model popover (Music, Sound Effects, Text to Speech or Transcribe) decides what the prompt box does; each Type remembers its own model.

What you can make

• Voice: type a script, pick a voice (a play button previews it; the first play is billed, later ones are cached) and get back speech. Cloud voices come from ElevenLabs, OpenAI, Google Gemini, MiniMax, Fish Audio, xAI, Kling, ByteDance’s Seed Audio and Microsoft’s MAI-Voice; Supertonic TTS runs on your machine with ten voices, and Phonon on an Apple silicon Mac. Fish Audio’s models, and HeyGen Voice on Runware, clone a voice from a sample you attach under Voice cloning.
• Music: describe a track; Music 2.6 and ACE-Step v1.5 add a Lyrics field, ACE-Step takes a Source audio clip to continue or restyle, and the Lyria family takes reference images.
• Sound effects: describe a sound and set a duration; SFX 1.5 also takes a Source video.
• Transcribe: select, Upload or drop a recording or a video (its soundtrack is extracted) and get a .txt, plus an .srt when the model returns timestamps. The transcript shows under the player with clickable timed lines, Copy and Open in Text. Whisper Base, Small and Large V3 Turbo transcribe on your machine.

Music and sound effects get the wand; speech is spoken verbatim, so it has none. A cost hint beside Create prices a clip, a script or a file before you make it.

Editing audio

The player draws the waveform with play, pause and a scrub bar (Space toggles playback). Trim with handles on the waveform, Fade in and out, Volume from silent to double, Speed from 0.5× to 2× (pitch shifts with it), and Format to MP3, WAV, FLAC or OGG. Every tool has a Save a new copy toggle, edits keep the file’s own format, and Info shows duration, format, size and the model and prompt behind a generated clip.

opening-shot.mp4

Veo 3.1 Fast

start frame

··· generated motion ···

end frame

0s

6s

The video workspace handles MP4, WebM, MOV and AVI files, one clip at a time: generating it, continuing it and cutting it. For stitching clips, audio tracks and titles into a longer video, see Compositions.

Generating video

Continuing and editing

• AI menu: Extend (use video) and Extend (use last frame) load the composer so you write what happens next; below them, Upscale to 1080p, Upscale to 4K and two background removals (green screen or transparent WebM), which run on Replicate.
• Trim, Resize, Crop, Speed (pitch preserved), Audio (extract an MP3 or strip the track), Background (the clip framed on a gradient, colour or picture), GIF and Format (MP4, WebM, MOV): all run locally through the bundled ffmpeg with a Save a new copy toggle.
• The transport has Loop, a preview speed and a camera button that saves the current frame as a PNG; arrow keys step a second, Shift+arrow a frame. Info shows duration, resolution, frame rate, codecs and the model and prompt behind a generated clip.

espresso-launch↶ ↷16:9 ▾Export☰

Helio launch

▶

0:03 / 0:09Timeline

0s2s4s6s8s

Title

Helio launch

Pre-order

Image

Hero

Generating…

Video

Demo clip

Audio

Voice + music bed

A composition is where single clips, images, audio tracks and titles become one finished piece. Where the video workspace produces one short clip at a time, a composition gives you a multi-track timeline and renders the whole thing to a single MP4, WebM or MOV. There are two kinds, fixed when you create one from the + in Compositions: Video (canvas and timeline) and Audio (speakers and tracks). Every composition is a file in your project folder.

Video compositions

Audio compositions

Pick Audio and you get a Script panel over a timeline with one lane per speaker plus Music and SFX. Each line is a voice clip on its speaker’s lane: add one with + Line, voice it per line or with Generate all, and later lines shift to fit once the real audio arrives. The Cast tab gives each speaker a name, a colour and a voice model with a preview; the chat drafts the whole script on request; export mixes everything down to MP3, M4A or WAV.

Promo thread

Llama 3.1 8B · local

Make a hero image of a coastline at dawn, 16:9

Saved · coastline-dawn.png

Now narrate the scene in a calm voice

ImageAudioVideoask in plain language, chat picks the tool

Chat is a single conversational workspace where you talk to a model and produce media in the same thread: ask questions, search the web, read documents and images you attach, generate and edit images, audio and video, or run a saved workflow. Everything it makes lands in your project folder as a real file.

Setting up a chat

What the chat can do

• Stream replies, with a collapsible Reasoning block and a step list that ticks off each model round and tool call. A turn chains up to six tool steps.
• Generate an image, a clip, a voice, music or an effect on request; edit the most recent file inline (“crop it to 16:9”, “convert to WebM”, matched locally before the model is called); update the last image; use two images as a video’s start and end frames; run a saved workflow by name.
• Read what you attach: any number of images on a vision-capable model, and PDF, text, Markdown or HTML documents, with a Digest whole document step for ones too large for a turn.
• Search the web with the Web toggle on (the default), from your own machine with no key, citing numbered sources under the reply; off keeps a chat offline.
• Regenerate a reply, edit and resend your last message, Stop a turn (what earlier rounds made stays), copy a message or the whole conversation as Markdown, and Read aloud a reply with your speech model.

A workflow is a visual diagram that chains models and edits together, a recipe you can run again and again. Great for repeatable jobs like “turn a blog post into a voiced video with subtitles”, or a batch over a whole folder.

The node types

• Input (a prompt or files), Model (runs an AI model of any modality), Edit (resize, crop, trim, convert, fade, speed, mute, extract, text overlay), Combine (merge files in parallel or series).
• Switch (you pick one of several inputs), Decide (a model picks, given a question), Bulk (a folder, a CSV or Input nodes become rows and every node after runs once per row).
• Preview (shows what flows through, saves nothing), File (saves into your project folder), Note (a label, never runs).

Building and running

The Assistant panel builds or changes a graph from plain language as one undo step, templates give running starts, Save as template keeps your own, workflows export and import as JSON, and Chat runs a saved workflow by name.

⌥-click or right-click a fileEdits, AI and CSuite

coastline.jpg

coastline.jpg

Open in CSuite

Convert›

Crop›

Remove background✦

Upscale✦›

Edit with a prompt↗

Run workflow↗›

1:1

4:5

16:9

9:16

CSuite can work on your files straight from your file manager, without opening the app. A small menu opens next to the file with everything that fits it: quick edits that run on your computer, AI actions that use your models, and actions that open the file in CSuite.

Opening the menu

• Right-click: on macOS, right-click a file in Finder, open Quick Actions and choose CSuite…. macOS adds new Quick Actions switched off, so turn it on once: choose Quick Actions → Customize and switch on CSuite…. On Windows, choose CSuite… in File Explorer’s right-click menu (on Windows 11, under Show more options). On Linux, Dolphin and Nemo list CSuite… in the menu, and Files has it under Scripts.
• ⌥-click or Alt-click: hold Option and click a file in Finder, or hold Alt and click a file in File Explorer or on the desktop, and the menu opens next to it once the click is done; a double-click doesn’t open it. The first time, macOS asks whether CSuite may read Finder’s selection. Linux has no click shortcut, because file managers there don’t share their selection with other apps.
• Open in CSuite sits on its own at the top. Below it the menu is split into edits, AI actions (marked with a wand) and other actions that open CSuite (marked with an arrow). A row with a chevron, such as Convert or Crop, opens its options when you click it; any other row runs when you click it. With the keyboard, the arrow keys move, Enter opens a row or runs an option, and Esc goes back. A click anywhere else closes the menu.

Edits on your computer

These run with ffmpeg on your computer, so they are free and need no account or provider.

• Images: Convert to PNG, JPG or WebP, Compress to 240p, 360p, 480p, 720p or 1080p, Crop to 1:1, 4:5, 16:9 or 9:16 from the centre, Rotate left or right, and Flip horizontally or vertically.
• Videos: Convert to MP4, WebM or GIF, Resize to 480p, 720p or 1080p, Compress to 240p up to 1080p, and for a video bigger than 10 MB also to 10, 25 or 50 MB (only the sizes smaller than the file are listed), Crop to 9:16 or 1:1, Rotate, change the Speed to 0.5× or 2×, Extract audio as an MP3, Save a frame from the middle as a PNG, and Smooth to 60 fps, which takes a while on long clips.
• Audio: Convert to MP3, WAV or M4A, Compress to 64 or 128 kbps, Speed 0.5× or 2× with the pitch kept, Pitch higher or lower at the same speed, Trim silence at the start and end, Even out loudness, Fade in and out, and Clean up voice, which lowers background hiss and evens out the level.
• PDFs and web pages: Extract text saves the text as a .txt file.

AI actions

AI actions use the models you chose in Models → Defaults (Text, and Transcribe and Text to speech under Audio, or Auto), need an active license, and show in Analytics like any other generation. Remove background and image Upscale use your Runware or Replicate account, and video Upscale uses Replicate. When something is missing, a notification says what to set up, and clicking it opens Models.

• Images: Remove background and Upscale 2× or 4×.
• Videos: Transcribe to subtitles (SRT or VTT) or to text, Burn in subtitles, Upscale to 1080p or 4K, and Summarize what is said. Burn in subtitles uses a .srt file with the same name next to the video when there is one, and transcribes the video first when there isn’t.
• Audio: Transcribe to text or SRT, Translate to a language you pick from the list, and Summarize. Translating a recording saves the translated text and a spoken version in that language, such as talk-es.txt and talk-es.mp3.
• PDFs, web pages, text and Markdown: Summarize into a Markdown file such as notes-summary.md. Text and Markdown files also have Translate to, Improve writing, which fixes grammar and clarity and keeps your formatting, and Read aloud, which saves an MP3 in your text to speech voice.

Subtitles need a transcribe model that returns timings. On Auto, CSuite picks one that does when you have one. Otherwise a notification says so before anything is uploaded, and clicking it opens Models so you can pick another under Audio → Transcribe.

Open in CSuite

• Open in CSuite: images, videos, audio and text files open in their workspace with the file selected, and the folder the file is in becomes your project folder, as if you had picked it in CSuite. Edit with a prompt and Trim open the file the same way. Formats the workspace can’t show, such as HEIC or MKV, are first converted to a copy next to the original.
• Edit with a prompt (images): opens the Image workspace with the image as the one to edit, ready for your prompt.
• Animate into a video (images): opens the Video workspace with the image as the start frame. If your video model doesn’t take a start frame, choose Auto or one that does, and the image is added.
• Trim (videos): opens the video in the Video workspace with the trim tool on.
• Send to Chat (images) and Ask in Chat (PDFs, web pages and text): starts a new chat with the file attached.
• Run workflow (any file): pick one of your saved workflows. The files go into its matching input and the workflow runs; when it has no input for that kind of file, it opens so you can add one.

Where the results go

Select one file or many. The menu only shows what works on every selected file, so an image and a song together offer only Open in CSuite and Run workflow. Each result is saved next to the original and never replaces it: photo.jpg becomes photo.png, a second run makes “photo 2.png”, and most actions add a word to the name, as in clip-720p.mp4, photo-1x1.png or photo-remove-bg.png. Compress and Resize set the shorter side of the picture, so portrait and landscape files both work; Compress never makes a picture bigger, while Resize does. Compressing a video to a size lowers its quality, and for long videos its picture size, until it fits; a video too long to fit at a watchable quality is skipped with a notice. Converting a file to the format it already has is skipped, and a converted JPG gets a white background where the original was transparent.

A notification tells you when the files are ready, and clicking it shows the file. AI actions and long video jobs also show one when they start. CSuite does not need its window open. With Keep CSuite running in the background on (the default on macOS and Windows), closing the window leaves CSuite in the menu bar or the notification area, so the menu opens instantly. If CSuite is not running at all, a right-click starts it in the background first. The first edit downloads ffmpeg once, which needs an internet connection. There is no cancel button, but quitting CSuite stops a running job and leaves no half-written file behind.

Account has six sections: Analytics, License, Instructions, Settings, Updates and Help. Provider keys live under Models, not here.

Analytics

• Four tiles head the panel: Estimated spend (with an “N unpriced” hint when some runs carried no rate), Generations, Tokens in / out and Avg est. cost / gen, over the last 7, 30 or 90 days, or all time.
• Below them: Estimated spend over time, Cost by model, Estimated spend by modality, Estimated spend by provider and Recent activity, so you can spot which workflow is eating your budget.
• The word estimated is doing real work: figures come from each model’s published rates, won’t match an invoice to the cent, and the actual billed amounts live on the provider’s own dashboard. The exceptions report the real billed cost, so those rows are exact: Runware text and Runware video (drafts, edits and upscales), everything through OpenRouter, xAI text and chat, and Black Forest Labs images.
• Local runs carry no rate, so they count as generations but add nothing to spend. They’re tracked alongside cloud usage so you can compare apples to apples.

License

• Buy a license: when the app is unlicensed, a button showing the live price opens the checkout in your browser. Until a key is activated the app opens and your files stay readable, but every prompt box is replaced by a License required panel, and workflow runs and composition exports are disabled with the reason in a tooltip.
• Activate a license: paste your key and click Activate. The app validates the key online once, then stores a signed receipt that works offline for 14 days. Both stay in the app’s local data on your computer.
• See your license status: active, active (offline) after a day without a check, offline grace expired, expired, revoked, or moved to another machine, along with when it was last validated. The app re-checks shortly after launch and every 12 hours in the background, and Refresh forces a check right now. If the key was activated on another computer since, re-entering it here moves it back and unlicenses the other machine.
• Deactivate a machine to free the license up for a different computer.

Instructions

Write your style or brand rules once instead of repeating them in every prompt. The panel has a Global box plus one each for Chat, Text, Image, Video, Music and Sound effects, each with its own on/off switch and room for 4,000 characters. They save as you type and apply from the next generation, including workflow Model nodes and the media Chat makes for you. Speech and transcription are left alone, since those work from your exact words, and so are the small assists (prompt enhance, ghost suggestions, selection actions, document digests). An info icon in the prompt box shows when instructions are being sent; click it to open the editor.

Settings

App-wide preferences, in one panel. Every change applies immediately, with no restart:

• Language: English, Spanish, Portuguese (Brazil), French, German, Italian, Japanese, Korean, Chinese (Simplified and Traditional), Russian, Hindi, Arabic, Bengali, Indonesian, Hebrew, Turkish, Vietnamese, Polish and Ukrainian, each listed in its own script. A fresh install follows your operating system’s language until you pick one here. Arabic and Hebrew flip the whole interface right-to-left.
• Appearance: System, Light or Dark (System follows your operating system; Dark is the default), plus a UI scale stepper from 100% to 140%.
• Notifications: an OS notification when a generation, workflow run or chat reply finishes, or fails. It only fires when you’re looking elsewhere, so it never interrupts you watching the thing you’re waiting for, and clicking it jumps to the screen that produced it.
• System integration: Show CSuite in the file right-click menu adds or removes the CSuite… entry, and Open the menu with ⌥-click in Finder (on Windows, with Alt-click in File Explorer) turns the click shortcut on or off. Both are on by default; see Right-click actions. Keep CSuite running in the background keeps CSuite in the menu bar (macOS), the notification area (Windows) or the tray (Linux) after you close the window, so its menus open instantly. It is on by default on macOS and Windows and off on Linux, where some desktops hide tray icons. Open at login starts CSuite in the background when you sign in, and needs the first switch on. The icon’s menu has Open CSuite and Quit CSuite.
• Privacy: a single switch, Share anonymous usage data, for the daily ping described below. Setting the DO_NOT_TRACK environment variable forces it off.
• Reset: returns the app to a fresh state. After a confirm dialog it relaunches and clears everything the app stores on the way back up: settings, provider keys, the license activation, downloaded runtimes and local models, and also your chats, workflows and saved templates, instructions and analytics history, which live in the app’s data rather than your project folder. You then set it up again from the welcome screen. Anything in your project folder is left untouched (composition files stay on disk too, but the list of them is cleared).

Updates

• CSuite keeps itself up to date: it checks for new versions shortly after launch, every six hours and whenever the window regains focus, downloads them quietly, and installs on the next restart. When an update is ready, a Restart to update button appears here (on macOS and Windows the same pill also sits in the title bar) and asks you to confirm before restarting, a small update button shows in the side menu while one downloads or waits, and an OS notification (“CSuite update ready”) opens this panel when you click it. The panel shows your current version and links to What’s new at csuite.so/updates.
• Prefer to stay in control? Untick Automatically download and install updates and use Check for updates whenever it suits you.
• The model list updates the same way. On the same schedule CSuite downloads the current list of models, and new ones appear the next time you open the app, with no new version needed. A model that needs a newer CSuite arrives with that update. Unticking the update setting stops this check as well.
• On Linux only the AppImage updates itself. A .deb install checks but doesn’t download (the toggle reads Automatically check for updates there) and shows “Version X is out” with a Download button, plus a “CSuite update available” notification.
• After three failed checks or downloads in a row, the app points you to the website to download the latest version manually.

Help

Help keeps every way to get unstuck on one screen. Under Guides and support, Open the guide opens this guide in your browser and the In-depth guides row opens the detailed page for each screen. Send an email starts a message to support@csuite.so in your mail app, and Open chat opens a live chat with the team in its own window inside CSuite.

Under Feedback and requests, Add feedback sends bug reports and feature requests straight from the app: a title and a description that takes Markdown. Each one shows when it’s marked done, and can come back with comments from the CSuite team as we follow up. Sending feedback requires an activated license.

The guide is never far away elsewhere in the app. An empty screen ends its description with a link to that screen’s in-depth guide, and on macOS the Help menu in the menu bar has CSuite Guide, Keyboard Shortcuts and Get Support.

The anonymous daily ping and its off switch live under Settings, above. See Privacy & data for exactly what it contains.

Every shortcut in one place. ⌘ is Ctrl on Windows and Linux.

CSuite is local-first. Here’s exactly where your data goes:

• Your files live in the project folder you picked. CSuite never uploads them to our servers. The only time a file leaves your machine is when you attach it to a cloud generation yourself (a reference image, a source clip, a chat attachment), and then it goes to that provider under your own key.
• Your prompts go directly to whichever provider you picked for that generation. Local providers stay on your machine; cloud providers receive the prompt through your own API key.
• Your provider keys are stored locally on your computer, encrypted at rest with your operating system’s secure storage where it provides one (Keychain on macOS, DPAPI on Windows, libsecret on Linux), never on our servers.
• Your license is verified online when you first activate it; after that the app uses a signed receipt that works offline and is quietly re-validated in the background.
• What the app tells our servers: license checks and feedback send only what’s needed to do the job (your license key, app version, platform, a machine identifier and, on activation, your computer’s name); the Buy button fetches the current price at startup; update checks just fetch the release feed and the current model list.
• Live chat stays off until you open it. Account → Help → Open chat loads the chat window from Crisp, our chat provider, which stores the conversation so we can reply. Nothing is sent to Crisp before you open that window.
• The anonymous usage ping: once a day, on by default, and a few more times during first-run setup, the app sends a random install id (unrelated to your license), the app version, OS, architecture, system and app language, and daily counts for the last seven days: generations that finished or failed per model, modality and provider type (a failure carries only a general cause such as “timeout”), where in the app they started, local model downloads, workflow runs, composition exports, and generations refused for lack of a license. It also sends how far you got through first-run setup, whether the app is licensed, which provider types you connected, the model catalogue version, and your memory and graphics hardware in broad ranges. A model you added yourself is reported only as “custom”. Our server notes an approximate country, never stores IP addresses, and keeps the data for 24 months. That is how we know which models people use and where the app fails them. The off switch is Share anonymous usage data under Account → Settings → Privacy, and the DO_NOT_TRACK environment variable overrides it.
• What we never see: your prompts, your generations, your project files, your chats, your API keys, and what any of it cost you in money or tokens.

A generation failed

• Check the model picker. Is the right provider selected?
• Open Models → Providers (or Models → Vendors for a lab’s own key), click the entry, and save the key again to re-validate it. Expired or invalid keys are the most common cause.
• “… refused the request: the account is out of credit or over its spending limit”: add credit, or raise the key’s limit, on that provider’s dashboard. “… the API key was rejected”: save a valid key under Models → Providers. Under Auto the generation has already tried your other platforms; this message means none of them could take it.
• Every prompt box replaced by a License required panel? Your license has expired or been revoked (or was activated on another machine). Open Account → License and press Refresh or re-activate.

A RunPod run failed

• “RunPod has no Secure Cloud GPU … free right now”: RunPod has no card with enough memory in stock. Try again in a few minutes.
• “Your RunPod balance is too low to rent a GPU”: add credit in your RunPod account, then run again.
• “…the OpenSSH client (ssh), which was not found on this computer”: CSuite reaches the GPU through the system’s SSH client. It ships with macOS and Windows 10 and later; on Linux, install the openssh-client package.
• “…CSuite could not remove it, so it may still be billing”: delete the GPU in your RunPod console (its name starts with csuite-), or CSuite removes it the next time you rent one. A GPU that never finishes starting also removes itself after 15 minutes.

Auto picked the wrong thing

• Auto reads the prompt with your Decision model, then falls back to your Text model. If neither is set up, pull a small Decision model from Ollama under Models → Providers; it’s fast and keeps the reading on your machine.
• Auto never picks vendor-only models, GGML-only models or models you added yourself. Pick those from the pill directly.
• A dialog saying Auto found a better model isn’t an error: continue with the model you have, or add the provider it names.

A local model is stuck or slow

• Bigger models need more memory. If your machine has limited RAM, try a smaller model from the catalog.
• Restart the app. Local runtimes free up memory on quit.
• An interrupted download isn’t lost: Ollama pulls and GGML weight downloads resume where they stopped once you’re back. If one is genuinely stuck, delete the model in Models and pull it again.
• Local video on Apple silicon is slow by nature. Wan 2.1 T2V 1.3B is the practical everyday pick; the larger models list the machine they need on their cards, so believe them.
• On Linux, the bundled Ollama runtime can’t install without zstd. Install it from your package manager (the .deb already depends on it).
• “Could not download the built-in Ollama runtime” or “Could not start the built-in Ollama runtime”: Ollama downloads wait for CSuite’s own Ollama and stop here rather than going to another Ollama on your machine. Check your connection and the Ollama system logs, then retry or restart the app.
• “Ollama is busy with a download or a generation. Update it when that finishes.”: Update to … waits for nothing to be running, since it restarts Ollama. Let the pull or generation finish, then press it again.
• “The Ollama server at … is too old to download …”: CSuite is set to use your own Ollama server. Update that server, or go back to the built-in one with Reset under Server URL in Ollama settings; for image models only Reset helps.
• On Windows, the HuggingFace runtime needs the Microsoft VC++ 2015–2022 Redistributable. Install it and retry the download.

A video edit was canceled or didn’t finish

• Quitting the app while a video edit is running cancels the job. Re-run it once the app is back open.
• Very large source files take time. The progress bar shows how far along the edit is.

A bulk run left some rows unfinished

• A node showing fewer rows than the total (“7 / 10”) had rows fail. Hover the count to see which ones and why, then press Run the rows that have no result yet on that node to retry only those.
• An error that starts with a row number, before anything has run, means a CSV row names a file that isn’t there. The number is the row your spreadsheet shows, with the header as row 1.
• A CSV needs a header row and at least one row below it. Spreadsheet files such as .xlsx aren’t read: export the sheet as CSV first.
• Only one row ran? Run mode is still set to Single (for testing out). Fewer than you expected? Raise the Limit on the Bulk node.

The right-click menu is missing or fails

• On Windows 11, CSuite… sits under Show more options at the bottom of the menu.
• On macOS, right-click a file and choose Quick Actions → Customize, then switch on CSuite…. It is also listed under System Settings → Keyboard → Keyboard Shortcuts → Services.
• Check that Show CSuite in the file right-click menu is on in Account → Settings, then start CSuite once: it adds the entry when it starts.
• ⌥-click does nothing: Finder must be the app in front, and the click switch on in Account → Settings → System integration. If you declined the macOS question, a notification says “To use ⌥-click, allow CSuite to control Finder in System Settings, Privacy & Security, Automation.” Clicking it opens that pane.
• Alt-click does nothing on Windows: File Explorer or the desktop must be in front, and CSuite must be running. With Keep CSuite running in the background on, CSuite stays in the notification area after you close its window.
• No tray icon on Linux: some desktops, such as GNOME without an extension, hide tray icons. CSuite keeps running if the setting is on; open it again from the app menu. On Wayland the menu opens in the middle of the screen with a Cancel button, because Wayland doesn’t let apps place windows at the cursor.
• On Linux, the entry appears in Files, Dolphin and Nemo. Thunar and other file managers are not supported.
• “CSuite doesn’t have permission to use this folder.”: on macOS, allow CSuite in System Settings → Privacy & Security → Files and Folders and run the action again.
• “No CSuite action works on all the selected files.”: the selection mixes types that share no action, such as an image and a song. Select files of one type and try again.
• “Couldn’t download ffmpeg.”: the first action downloads ffmpeg once, so connect to the internet and try again.
• “Set up a text model in CSuite first.”, or the same for a speech-to-text model or a text-to-speech voice: none of your providers or installed models can run that part. Click the notification to open Models, then add a provider or pick a model under Defaults.
• “This needs a Runware or Replicate account in CSuite.”: Remove background and Upscale run on those providers only. Add a key under Models → Providers.
• A row you expect is missing, such as Smooth to 60 fps, Even out loudness, Clean up voice or Burn in subtitles: the copy of ffmpeg on this computer lacks the part that action needs. The menu also hides actions that don’t fit every selected file.
• “No speech was found in this file.”: the recording has no words the transcribe model could hear, so nothing was saved.
• “No text was found in this file.”: the PDF is a scan, an image of pages with no text in it, so Extract text has nothing to save.
• “Couldn’t open photo.jpg in CSuite.”: files in hidden folders or in CSuite’s own data folder can’t be brought in, and a chat attachment must be under 100 MB. Move the file somewhere else and try again.

The app is in the wrong language

• A fresh install follows your operating system. Pick the one you want under Account → Settings → Language; it switches on the spot.

Still stuck? See Support below for how to get in touch.

CSuite is built and supported by a small team. Every email goes to a real person, and we read every one.

• Email support@csuite.so for bugs, questions, license help, and feature requests. Inside the app, Account → Help has Send an email, and on macOS so does Help → Get Support in the menu bar.
• Chat with us live: Account → Help → Open chat opens a chat with the team in its own window inside the app.
• Send feedback without leaving the app: Add feedback is on the same screen.
• Include your operating system, a short description of what you were doing, and (if relevant) the provider and model you had selected. The more we know up front, the faster we can help.
• Lost your license key? Email us from the address you used to purchase and we’ll resend it.
• Have an idea? We genuinely want to hear it. The same inbox is the right place.

CSuite is a one-time purchase, sold with a no-questions-asked refund window. We’d rather you spend money on something you love than feel stuck with something you don’t.

• 7-day refund: if CSuite isn’t the right fit for any reason, email us within 7 days of purchase and we’ll refund your order in full.
• No forms, no surveys: a one-line message is enough. We don’t require a reason, though we’re always grateful when people share one so we can improve the app.
• How it works: refunds are issued to the original payment method via Stripe (typically 5 to 10 business days to land back on your statement). Your license key is deactivated as part of the refund.

To request a refund, email support@csuite.so with the email address you used at checkout.