How can we help?

Convira is a desktop AI agent that can research, organize files, use a visible browser, and create or edit Word, Excel, PowerPoint, and PDF documents. You can watch the work, approve sensitive actions, and choose where the AI processing runs.

Local AI runs an Ollama-compatible model on your machine - your context, prompts, memory, and files stay on the device. The one exception is a billed tool you invoke (web and X research, image and video generation, connector actions): it needs managed provider keys we hold, so that single call runs on Convira when you're online and is unavailable offline. Best for sensitive work and offline use.

Cloud uses hosted AI models with no local model setup. Convira processes the request and forwards the necessary prompt and context to the selected model provider, but does not retain the run content in its database. The authoritative run record stays on your device.

Private Box is a strong machine you control on your LAN. For a run you start, the desktop talks to it directly and the prompt does not pass through Convira; a task you arm to run while your computer is off is the exception, because arming means Convira runs it for you and sends it to the box. Convira licenses the box with a short-lived signed credential your desktop renews, within a window the box's owner sets, so it needs an active subscription and periodic connectivity. Good for bigger models that don't fit on a laptop without leaving your network.

Any Mac with Apple silicon (M1 or newer) running macOS 12 Monterey or later.

Windows 10 and Windows 11 on 64-bit Intel and AMD processors (x64). One installer covers both.

Ubuntu 22.04+, Debian 12+, and Fedora 38+ on x86_64. The .deb installs on Debian and Ubuntu; the .tar.gz unpacks and runs on any other distro.

Convira checks for a new version on launch and every so often while the app is open. Nothing downloads until you ask it to. On macOS and Windows the update installs when you restart or quit, and those builds also use the operating system's code-signing checks. On Linux the app tells you a new version exists and links you to the download. Update traffic is skipped when Convira's egress policy blocks it.

Convira processes the request and forwards the necessary prompt and context to the selected AI model provider. Run content is not written to Convira's database; the authoritative session record remains on your device. The one exception is a chat you choose to share when you rate an answer, which we keep for up to 90 days, or until you remove the rating. Operational metadata such as status, model, tools, token use, and credit use may be retained. See the Security and Privacy pages for the full data-path explanation.

Convira supports its curated local catalog and compatible Ollama models. The model picker checks your available hardware and shows which choices fit, so you do not need to compare model sizes by hand.

The same catalog as Local AI plus much larger models - from 119B up to trillion-parameter mixture-of-experts models - that no laptop can hold. Because the box has memory a laptop doesn't, Convira allows bigger models there.

Yes - a Private Box is shared across your workspace. A teammate with Manage settings seals a separate per-member key to each member's device key (X25519), so Convira never sees it in plaintext. Each member's desktop unseals its own key locally to reach the box.

Cloud and Private Box need network reachability. Local AI runs offline once you've pulled the models - for up to 24 hours between check-ins. There's a one-click Offline Mode toggle in the app footer. It blocks all outgoing traffic except your machine's own local services and Convira's sign-in and licensing endpoints, and cuts off in-flight requests the moment you flip it.

To run Convira: 8 GB of memory and about 4 GB of disk. The app bundles its document engine and the small models it uses to search your files and to type what you say.

To run AI on your own computer: about 12 GB of memory fits one of the smaller models and about 16 GB fits a bigger one. Above that, more memory first buys a longer conversation memory, then a larger model. With 8 GB the cloud engine is the better choice, and you can still run a model on your own computer if you want to. Models download from about 3 GB to about 76 GB of disk each, and the ones that fit a laptop are at the small end. Device check gives the exact answer for your computer, and the model picker in the app only offers what fits.

Device check answers that before you install anything: name your computer or enter its parts, and it shows which models fit, how much conversation memory each one gets, and when the cloud engine is the better choice. It works from the same fit rule the app uses, so the model picker inside Convira agrees with it.

Plans differ by included cloud credits, collaboration features, and Private Box access. See the Pricing page for the current plans and billing options.

macOS: drag Convira from Applications to the Trash.

Windows: Settings → Apps → Convira → Uninstall.

Linux .deb: sudo apt remove convira-desktop. Linux tarball: delete the unpacked folder.

Use the form on this page - the message field lists what we need. The one thing to copy exactly is the error text from the wizard, since that usually points straight at the cause.