Personal Jarvis
Coordinate coding agents, voice assistance, and computer actions from one desktop app.
- Source repo
- PersonalJarvis/PersonalJarvis
- Stars
- ★ 157
- Last updated
- today
- License
- Apache-2.0
- Primary language
- Python
- FA score
- 71/100 · Some gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code · OpenAI API · Claude API
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A developer comparing approaches can assign the same coding task to several Agentic IDE terminal panes and review the results.
- Not a fit if
- Users unwilling to approve desktop actions or review activity logs
- macOS users who need an app image notarized by Apple
- Source review
- 71/100 · Some gaps
What does this agent do, and when should you use it?
Personal Jarvis is an open-source desktop app for Windows, macOS, and Linux, and it can also run as a headless server. Its Agentic IDE runs multiple coding agents side by side in real terminal panes, while Jarvis Agents provide persistent specialists with their own chats, tools, memory, and schedules. Users can talk through a wake word or shortcut, dictate into apps, and let Jarvis operate the desktop and browser. It supports multiple model providers, local models, and local speech options, alongside a plugin marketplace, MCP connections, a memory wiki, and scheduled routines. Actions that could change the system require approval, runs are recorded, and users need an API key, subscription, or local model.
Users can open a project and launch installed coding tools such as Claude Code, Codex, or OpenCode in Agentic IDE terminal panes, send the same task to multiple panes, and follow their work. Through voice or text, Jarvis can route a task to a pane by call sign or to a Jarvis Agent; agents use their configured tools and memory and save files, reports, and plans in Artifacts. Jarvis can also read and operate the desktop or browser, dictate into other apps, connect marketplace plugins or custom MCP servers, and run scheduled routines. Its local wiki stores and cites information across sessions. Actions that could change the system require approval, and every run is recorded.
- A developer comparing approaches can assign the same coding task to several Agentic IDE terminal panes and review the results.
- A solo worker handling recurring research, release, or marketing tasks can create Jarvis Agents with their own tools, memory, and routines.
- Someone who types across desktop apps can dictate by voice and issue tasks using a wake phrase or keyboard shortcut.
- A user with recurring workflows can schedule a morning briefing, inbox sweep, or nightly report.
- A user who needs an agent to inspect or operate the desktop or browser can enable computer use and approve actions that may change the system.
How do you install or deploy this agent?
The installer checks for Python 3.11 or newer and Git, and offers to install anything missing. After installation, choose a language, wake phrase, and model in the app. You need an API key, a subscription, or a local model; a microphone helps with voice features, but no GPU is required.
Windows (PowerShell):
irm https://raw.githubusercontent.com/PersonalJarvis/PersonalJarvis/main/install/install.ps1 | iexmacOS and Linux:
curl -fsSL https://raw.githubusercontent.com/PersonalJarvis/PersonalJarvis/main/install/install.sh | bashDesktop installers are also available for Windows, macOS, and Linux. The macOS disk image is not yet notarized by Apple; the first launch requires allowing it in Privacy & Security, with steps varying by macOS version. For server deployment, install the package with pip.
How do you use this agent?
Complete the in-app setup by choosing a language, a wake phrase or keyboard shortcut, and a model. Then say the wake phrase and request a task, open a project folder in the Agentic IDE, or create an agent.
jarvisThis starts the full desktop app. To run it as a server, install the package and start the service, then open the local address in a browser:
pip install personal-jarvis
jarvis serveVisit http://localhost:47821. Microphone use from another computer requires HTTPS. Connected services may require their own setup, and computer actions may require approval.
What are this agent's strengths and limitations?
- Agentic IDE runs multiple coding agents in real terminal panes and lets users route spoken instructions by short call sign.
- Jarvis Agents retain separate chats, tools, memory, and schedules, and save their outputs for review.
- The app supports several hosted model providers, Ollama, and OpenAI-compatible local servers; speech recognition and synthesis also have offline options.
- It documents desktop support across Windows, macOS, and Linux, plus headless server deployment and desktop/browser control.
- Using a model requires an API key, subscription, or local model; paid providers or subscriptions may add costs.
- The automated installer depends on Python 3.11 or newer and Git; server microphone access from another computer requires HTTPS.
- Actions that may change the system wait for approval and runs are recorded, which may not suit users seeking unattended system control.
- The macOS disk image is not yet notarized by Apple, so first launch requires a manual approval step.
How does this agent compare with similar options?
The README presents Claude Code, Codex, OpenCode, and other coding agents as tools that Personal Jarvis can run side by side, rather than as products it replaces. It also supports Claude and ChatGPT subscriptions. The provided material does not compare it feature by feature with other desktop assistants.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Personal Jarvis This agent | 71 · Some gaps | Desktop appFree + model costs | ★ 157 | today | Python | Codex · Claude Code · OpenAI API · Claude API |
| AI DevKit | 48 · Major gaps | CLIFree + model costs | ★ 1.6k | today | TypeScript | Codex · Claude Code |
| Coder Eval | 86 · Good | CLIFree + model costs | ★ 151 | 2d ago | Python | Codex · Claude Code · OpenAI API · Claude API |
| Runner Multi-Agent Terminal | 82 · Some gaps | Desktop appFree | ★ 275 | today | Rust | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The security policy describes safe, monitor, ask, and block tool tiers, approval for actions that change the system, OS credential storage, refusal to accept secrets in voice and chat, rollback for configuration changes, isolated worktrees, and audit records. These are specific design claims, but the supplied material does not include implementation evidence. The app can also operate the screen, keyboard, command line, phone calls, and connected services, so least privilege, data flow, and external effects earn adequate rather than full scores. Attribution is limited to README and license details; publisher identity remains unknown.
The README, security policy, dependency manifest, and CI workflow broadly align on capabilities, platforms, and optional dependencies. The manifest sets dependency floors; the canary describes cross-platform runtime checks; and missing optional components are said to degrade gracefully or produce errors. Static evidence cannot confirm that every implementation matches these claims, so scores are not full.
The materials address desktop users, developers, and headless servers, with concrete scenarios for voice, coding agents, browsers, tools, models, and multiple operating systems. They also explain platform-specific dependencies and some fallback behavior. The scope is broad, but automatic skill selection is described only as matching a request, without sufficiently precise trigger conditions or boundaries.
The README has substantial navigation, installation steps, platform notes, troubleshooting links, and feature guides. SECURITY and the dependency manifest add supported-version policy, reporting routes, limitations, and platform distinctions. The license text and package version are clear, including the MIT licensing of releases through 1.6.0. The supplied material lacks changelog contents and does not fully establish ongoing maintenance responsibility or stable naming conventions, so those criteria do not receive full marks.
The product combines coding agents, persistent agents, voice, memory, routines, and computer control in one local application. Artifacts, tool records, and visible steps suggest concrete integration value. Output quality and cost-benefit are not established by runtime evidence; the broad feature set also requires users to configure models, services, and permissions, so these scores remain moderate.
The README links to security, privacy, installation, architecture, and feature documentation; SECURITY, pyproject.toml, and the workflow provide traceable architecture claims, dependency constraints, and CI design. Some claims receive support across files, but key safety and capability claims remain largely project-authored and runtime behavior cannot be independently verified from the supplied material. Full marks are therefore not justified.
- The app can read screens and control desktop input, run command-line tools, place calls, and use connected services. Review its safety and permission settings, and heed its documentation's advice to protect API keys and sensitive data.
- Security controls and cross-platform behavior were assessed from static materials only; their runtime behavior is not established here. The macOS disk image is not yet notarized, Linux global shortcuts require an extra install, and Wayland does not support this kind of global key grab.
FAQ
Is Personal Jarvis free to use?
Does my data leave my computer?
Can Jarvis control my computer without asking?
Can I use it entirely offline?
How can I run it?
jarvis serve.