Frequently asked questions
How Vivianne works, what it runs on, and what’s coming next.
A language model on its own only produces text. An agent harness is the runtime around it: the loop that plans and calls tools, manages context and memory, asks for approval before risky actions, and checkpoints work so it can resume. Vivianne is a harness you install on your computer, so any capable model becomes an agent that finishes real tasks.
Yes. Two people pair their Vivianne apps with a short code over a direct, encrypted peer-to-peer connection, follow each other’s agent sessions turn by turn, and chat to divide up the work, with no account or central server required.
Yes. Coding is what it does best today: it reads and edits your codebase, runs commands in a real terminal, delegates big jobs to subagents, and shows every file change for review. Math, science, and research workflows are being built on the same harness.
Yes. It writes and runs code to check calculations, analyses data, and produces real documents. Science work is live too: chemical structures and reactions, 3D molecules and proteins, biology diagrams, physics graphs and interactive diagrams, with Greek letters and chemical equations anywhere.
Anthropic, OpenAI, Google, and a dozen more providers, plus any OpenAI-compatible endpoint. You can switch models mid-session, or run a local GGUF model entirely offline.
Yes. Point Vivianne at a GGUF model file on disk and it runs inference itself, with no API key and no network connection required.
Windows and macOS. Windows ships as .exe and .msi installers; macOS ships as one universal .dmg for Intel and Apple silicon. Linux builds are on the way.
Something else? hello@vivianne.dev