Local-first, encrypted
Runs on your hardware, with a model on your own machine or a cloud LLM under your own key — there's no Aivyx PA server in the path. Storage is encrypted at rest with a passphrase only you hold.
Aivyx PA — an agent that's actually yours.
Most assistants are someone else's service you rent. Aivyx PA is software you run — with the security model to make trusting it reasonable.
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/Aivyx-Agent/aivyx-pa/releases/latest/download/aivyx-cli-installer.sh | sh
aivyx-pa init # finds your model, sets it up, checks it worksWhat it does
It learns how you work
You decide who it is when you set it up. As you work together it learns your preferences and habits, and any change to its character waits for your approval. It becomes yours.
Permissions and a record
Everything it does is checked against what you've allowed, and written to a record that can't be quietly changed — you can verify it yourself, offline. It can't reach further than you let it.
A team when you need one
For bigger jobs, a lead assistant splits the work between specialists, each allowed only what its part needs, and checks the result. Local today; designed to span machines tomorrow.
Autonomy you dial in
One setting, from asking before everything to working on its own. It only gets more freedom when you give it — it can't raise its own limits or act unattended past the line you set.
Your model, your tools
Use a model on your own machine (Ollama, Lemonade, llama.cpp, Jan or Docker Model Runner) or Anthropic and OpenAI. Connect Gmail, Calendar and more, or any MCP server. One program, no signup.
"The agent gains autonomy by your choice — never one it grows into on its own."The line that makes trusting an autonomous agent reasonable.
More on Aivyx PA
Security, teams, and identity — the ideas behind it, in more depth.