Dodaios is an open-source agent OS for your desktop: it runs your coding agents, browser and apps by voice or chat, on your own machine. Formerly Personal Jarvis.
Python agent-orchestrationai-agentai-coding-assistantclaude-codecoding-agentcomputer-usedesktop-applocal-firstmcppersonal-jarvisvoice-assistantvoice-coding
docs(readme): explain the voice path, GPT-Live and the local realtime server
rubenluetke10-beep2c84130477b0
diff --git a/README.md b/README.mdindex acff92e36a..5f912264bb 100644--- a/README.md+++ b/README.md@@ -13,11 +13,13 @@Personal Jarvis is an open-source AI assistant that runs on your own computer and that you simply talk to. Say your wake phrase and it answers out loud, works your browser and desktop apps, and runs your coding agents for you: Claude Code, Codex, OpenCode, Kimi and others sit side by side in one workspace, and you can tell any of them what to do by voice ("tell T1 to run the tests"). One app for Windows, macOS and Linux, or headless on a server.+**Voice first.** Talking is the main way you drive it. The wake phrase is detected on your own machine, and then Jarvis opens a live two-way audio session: you can interrupt it mid-sentence, and it keeps the conversation going while the actual work happens behind it. With [GPT-Live](https://github.com/PersonalJarvis/PersonalJarvis/blob/main/docs/gpt-live.md), OpenAI's realtime voice carries the conversation over WebRTC while a separate thinking model does the reasoning and the tool calls, both on the same OpenAI key and with cloud recording switched off. Gemini Live is there too. If you'd rather keep your voice at home, Jarvis installs and manages a local realtime server on your own GPU (12 GB and up) that listens, thinks through Ollama and speaks entirely on your machine, and stays warm so the next call connects right away. Whichever you pick, every action goes through the same approvals, runs exactly once even if the connection drops, and keeps going after you hang up.+**A team of agents, not a single chatbot.** Jarvis is the one you talk to, and behind it you build a team. Each agent has its own name, instructions, memory and browser, runs its own routines on a schedule, learns from what worked in its past runs, and stops to ask you a multiple choice question when it isn't sure. You can watch them work in a small office you can walk through, and move any of them onto another computer over SSH. Coding work gets the same treatment: in the [Agentic IDE](https://github.com/PersonalJarvis/PersonalJarvis/blob/main/docs/product/extend-and-automate/agentic-ide.md) every coding agent has its own pane and, if you want, its own git worktree, and Jarvis always knows what each one is doing ("what is T2 working on?").**It conducts your tools instead of replacing them.** Most AI apps want you to move into their chat and their subscription. Jarvis sits on top of what you already have: your Claude Code or Codex plan, one API key from any provider, or a local model through Ollama. It turns them into a single assistant you can talk to, and you can swap any of them without touching anything else.-**How it's built.** A Python core and a React desktop app, both running locally. A small router only decides where a request goes (answer, tool, or agent) and every action passes through one executor with a risk policy (safe, monitor, ask, block), so nothing that could change your system happens without your approval, and every run is recorded. Voice streams end to end, so Jarvis starts answering before it has finished thinking. State, memory and keys stay on your machine; there is no account, no paid tier and no analytics. The whole design is in the [architecture overview](https://github.com/PersonalJarvis/PersonalJarvis/blob/main/docs/architecture-overview.md).+**How it's built.** A Python core and a React desktop app, both running locally. A small router only decides where a request goes (answer, tool, or agent) and every action passes through one executor with a risk policy (safe, monitor, ask, block), so nothing that could change your system happens without your approval, and every run is recorded. State, memory and keys stay on your machine; there is no account, no paid tier and no analytics. The whole design is in the [architecture overview](https://github.com/PersonalJarvis/PersonalJarvis/blob/main/docs/architecture-overview.md).[Website](https://personaljarvis.ai) · [Docs](https://github.com/PersonalJarvis/PersonalJarvis/blob/main/docs/product/start-here/welcome-to-personal-jarvis.md) · [Getting started](https://github.com/PersonalJarvis/PersonalJarvis/blob/main/docs/product/start-here/first-run-setup.md) · [How it works](#how-it-works) · [Discord](https://discord.gg/x7USduHxbc) · [X](https://x.com/PersonalJarvis)