The agent harness for code, math, and science.
One desktop app for your own agent, coding agents over ACP, and assistants that keep working. Any model, your keys.
One desktop agent for your code, your computer, and your browser, with any model.
24/7 agents
Always-on assistants with apps, browser, screen, memory, and schedules. Approvals show up inline.
Computer controlWindows
UI Automation with a pixel fallback. A notch shows activity and a Stop button.
Any model
Major clouds, any OpenAI-compatible endpoint, your ChatGPT plan, or local llama.cpp.
Tasks that survive
Waits out rate limits, falls back on failing models, retries, and resumes after a restart.
Cost-aware routingBeta
viv-auto picks the model for each step to keep cost down, with usage tracked by model.
MultiplayerExperimental
Pair two Vivianne apps and see each other’s agents work, live.
Memory and new tools
Recalls your instructions and facts. When no tool fits, it writes, tests, and keeps one.
Coding agents over ACPIn development
Install agents like Claude Code once, then run them in Vivianne.
Also built in: plan mode, a browser extension, MCP, skills, mods, and Git review.
What’s new
Full ChangelogOne harness. Every side of the work.
Windows and macOS today. Linux is on the way.