jarvis vs agent-starter-node
Two of the top projects in Voice and realtime, side by side: who leads on each score part and who each one suits, then setup, license, activity and what each review found.
jarvis
Voice interface that brainstorms specs and drives Claude Code sessions on macOS
agent-starter-node
Node.js voice AI agent starter for LiveKit Agents
Verdict
jarvis ranks higher, 58 to 54. Of the 4 score parts we measure for both, jarvis leads on 1, agent-starter-node on 2 and 1 is even.
- Adoptionjarvis leads by 26 points, 41 to 15: a wide gap.
- FreshnessEven at 100.
- Maintenanceagent-starter-node leads by 2 points, 40 to 38: close to even.
- Agent-readyagent-starter-node leads by 30 points, 70 to 40: a wide gap.
Pick jarvis if you are
Claude Code users on macOS who want to drive builds by voice
- Runs on the Claude subscription login; strips ANTHROPIC_* env vars so no API key is billed
- Watches all local Claude Code sessions and can answer a Terminal permission prompt
Pick agent-starter-node if you are
Node.js developers building voice agents on LiveKit
- Models via LiveKit Inference need no separate provider API keys, including realtime models
- Turn detector covers 14 languages; adaptive interruption handling and noise cancellation included
| What we compare | jarvis | agent-starter-node |
|---|---|---|
| Score parts, out of 100 | ||
| Adoption | 41, known | 15, niche |
| Freshness | 100, active | 100, active |
| Maintenance | 38, patchy | 40, patchy |
| Agent-ready | 40, minimal | 70, partly |
| Facts from GitHub and the README | ||
| Stars | 978 | 116 |
| License | custom license (read the license) | MIT (permissive) |
| Last commit | Sep 2026 | Nov 2026 |
| Last release | Sep 2026 | None published |
| Language | Python | TypeScript |
| Docker | No | Yes |
| GPU | Not needed | Not needed |
| arm64 or Apple Silicon | Not stated | Not stated |
jarvis
JARVIS is a voice front end for Claude Code on macOS. A long-lived `claude -p` process on your own subscription brainstorms a design with you, writes it to a spec file in the project, then starts Claude Code runs that plan, review and execute it. It also watches every Claude Code session on the machine and says aloud which ones are waiting on you. Runs are logged in SQLite and shown on a local dashboard.
Who it is for: Claude Code users on macOS who want to drive builds by voice
Strengths
- Runs on the Claude subscription login; strips ANTHROPIC_* env vars so no API key is billed
- Watches all local Claude Code sessions and can answer a Terminal permission prompt
- Every run is recorded in SQLite with prompt, status, tokens and events; six-tab dashboard
- Extra tools via any MCP server in connections.json; other MCP configs are ignored
Weaknesses
- macOS only; terminal control, screenshots and notifications depend on AppleScript
- Requires Chrome for Web Speech API; no server-side transcription fallback
- Fish Audio API key is required; without it replies appear only as text
- Spawned runs use --dangerously-skip-permissions by default, with full privileges on your machine
- Python, Claude (Sonnet by default, via Claude Code)
- Needs Claude Code CLI 2.1.224+ (logged in), Fish Audio API key, Google Chrome, Python 3.11+, Node.js 18+, Playwright Chromium, OpenSSL (self-signed certs)
- env example file
agent-starter-node
TypeScript starter for a voice AI assistant built on LiveKit Agents for Node.js. It wires up a speech pipeline through LiveKit Inference (default LLM Gemma 4 31B, default TTS Fish Audio S2.1 Pro), with turn detection, interruption handling and noise cancellation. It also includes simulation tests that run in CI and a Dockerfile for LiveKit Cloud deployment.
Who it is for: Node.js developers building voice agents on LiveKit
Strengths
- Models via LiveKit Inference need no separate provider API keys, including realtime models
- Turn detector covers 14 languages; adaptive interruption handling and noise cancellation included
- Simulation scenarios in scenarios.yaml run in GitHub Actions on every merge to main
- Ships AGENTS.md, a CLI debugger and a Dockerfile; pairs with React, Swift, Flutter, Android starters
Weaknesses
- Defaults assume a LiveKit Cloud account; self-hosting means swapping in model plugins
- Noise cancellation plugin is LiveKit Cloud only and must be removed when self-hosting
- Simulations use real inference on LiveKit Cloud, so CI runs cost money
- No frontend included; web or mobile client lives in a separate repo
- TypeScript, Gemma 4 31B, Fish Audio S2.1 Pro, OpenAI, Cartesia
- Needs LiveKit Cloud, LiveKit CLI 2.18.8+, pnpm
- GitHub template
- Docker
- env example file
More in Voice and realtime
Building with these?
We help teams pick an open-source AI stack and run it in production. Client work never changes a rank.