deepnovacore/NovaPersonalAgent

English | 简体中文

An open-source personal agent with private memory, smart proactivity and realtime interaction.

Nova draws on your context and private memory to understand what matters to you. It knows how and when to clarify, ask for approval, and move work forward. Its dual brains keep the conversation responsive while background agents handle longer tasks. Refine your goals as work progresses, revisit the results, and stay in control.

Website · Documentation · Video demo

Quickstart · Highlights · Use cases · Roadmap

• 2026-10-13 · v0.4.0 — Continuous context, ongoing work: GUI-use across browser, computer and phone (experimental), Scheduled Agent Tasks, and a full Timeline for context, evidence and task results.
• 2026-10-07 · v0.3.0 — Your personal agent: Todos, Ideas, Goals, Feeds and Profile, with source-backed memory and evidence-checked tasks.
• 2026-09-21 · v0.2.0 — Voice, your way: configurable ASR / LLM / TTS, personal memory, MCP and an iPhone companion.
• 2026-08-31 · v0.1.0 — Always-on voice assistant with dual brains: responsive frontend conversation and background agent execution, with live steering.

Your context informs the conversation; the front brain coordinates work with the back brain. Nova checks results and speaks when it matters.

• Your proactive personal agent. Todos, Ideas, Goals, Feeds and Profile turn what Nova knows about you into suggestions you can act on.

• Wide and unstructured context. Bring folders, mail, calendars, conversations and documents; keep personal memory traceable, editable and removable.

• Always-on voice assistant with dual brains. The front brain stays in conversation while the back brain handles long-running work.

• A thin frontend task coordinator. Clarify, authorize and delegate to a coding agent; check evidence against your criteria, take over and hand control back.

• Restrained proactivity. Nova chooses what matters and when to speak. Routine updates stay quiet, and reminders respect your turn.

• More coding agents. New sessions can also run on OpenCode, CodeBuddy, Pi or DeepSeek Harness over ACP; each session keeps the agent it started with. Live steering stays a Codex feature. See Coding.

How Nova works · Task controls · Source permissions

Connect the sources you choose: folders, mail, calendars and conversations. Personal memory and imported document knowledge help Nova understand your work; reading sources and model processing remain separate permissions.

Your personal workbench Turn scattered context into source-backed recaps and suggested todos. Choose what to do next, then continue with Nova.	Profile and personal memory Nova builds an editable picture of your work and interests. Trace personal memories to their sources, correct them or remove them.
Ideas and goals Keep an idea, choose a direction, and turn it into a todo when you are ready. Suggestions wait for you to adopt them.	Feeds that follow your interests Review public news ranked by your Profile. Save a story or turn it into an idea, todo or goal.

The front brain handles conversation; the back brain does the work. Task events and camera observations inform optional updates, while speaking has its own timing.

Understands what you mean Describe your goal naturally. Nova asks for missing details before starting a task; keep talking and refining it while the back brain works.	You set the boundaries Nova coordinates tasks and checks results against your criteria. Review permissions, take over when needed, and hear updates when they matter.
Voice-run workspaces Create and switch workspaces and sessions by voice. A new workspace waits for your confirmation.	Bring your tools and knowledge Configure ASR / LLM / TTS and MCP; ask questions across your documents.

Voice Vibe Coding Describe what you want, let a coding agent build and test it, and refine the work through conversation.	Camera Monitoring Ask Nova to watch for a condition and tell you when it happens. Observation runs independently of coding tasks.
Weekly Report Ask Nova to recall recent work and help organize a weekly report. Watch the original phone demo. ▶ Watch the demo	Take Nova with you Connect your iPhone over Tailscale to your desktop or headless server. Talk to Nova and approve tasks wherever you are.

Quick downloads · v0.4.0 — Release notes and SHA-256 checksums · Latest release

Requirements: Node.js 22.13+ and npm. For coding tasks, install and authenticate your chosen coding backend; Codex uses its app-server transport.

Install the release with npm.

npm install --global nova-personal-agent@latest
# open the shipped app; configure a provider to enable voice
novapersonal
# open the settings panel in the app
novapersonal config
# list the keys your voice pipeline needs; --online tests them
novapersonal doctor

v0.4 introduces the Nova Personal Agent package names. Published v0.3.x versions retain the old package names and novaaudio command. The new package keeps that command alias and reuses existing data and settings. When migrating an existing global install, first run npm uninstall --global nova-audio-agent nova-audio-agent-server to avoid command-link collisions; this leaves user data intact.

Quit Nova before upgrading, then rerun the install command above. To pin a version, use nova-personal-agent@0.4.0. Upgrades keep your settings and data; back up your Nova data before downgrading.

Headless Ubuntu 22.04+ x64:

1. Install or upgrade with npm install --global nova-personal-agent-server@latest, then configure the service and initialize credentials using the configuration guide.
2. Start the service with novapersonal-server start.
3. In a second terminal, run novapersonal-server pair wss://your-host.ts.net to display a one-use pairing QR code.

For development from source:

git clone https://github.com/deepnovacore/NovaPersonalAgent.git nova-personal-agent
cd nova-personal-agent
npm ci && cp .env.example .env

Get an API key from DashScope and set DASHSCOPE_API_KEY. That one key runs voice, memory, the camera and web search. Search goes through Bailian until you add a Tavily TAVILY_API_KEY.

npm run start:client
# Open Workbench for this launch, overriding the saved startup preference
npm run start:workbench
# Build, diagnose or run the local demos
npm run build --workspace @nova-personal-agent/runtime
node runtime/dist/src/cli.js diagnose --json
node runtime/dist/src/cli.js demo all

The client includes microphone, camera, sound, settings and external MCP controls. Try hovering over the desktop orb for a surprise :)

Models, credentials and per-provider limits: support matrix.

• v0.4.0 — Ongoing context and specialist agents. Persistent home and Timeline, scheduled tasks, incremental Profile updates and multiple coding backends. GUI/device execution remains experimental.
• v1.0.0 — Stable release. Establish stable core behavior across supported platforms, reliable installation and upgrades, predictable recovery, and complete user documentation.

Development integrations are not release guarantees. Features ship after their supported configurations pass acceptance.

npm ci && npm run check && npm run build && npm test

Live integrations are credential- and hardware-dependent and never substitute for the deterministic tests. Security reports: SECURITY.md; contribution rules and invariants: CONTRIBUTING.md.

Copyright 2026 DeepNovaCore, Apache License 2.0.