Multi-sensory feedback for AI coding agents. When your coding agent finishes a task, someone picks up the phone and tells you — cheerfully if it went well, furiously if it didn't. Hold music while it works, a cast of cloned voices when it's done, and a buzz on your mouse or a flash on a pixel display to match — all 100% on your own machine.
"Not happy, Jan — the issue is…"
Built on the agent-feedback idea from AgentVibes — and taken way too far.
📖 Read the docs · 🎬 Watch the demo (videos coming soon)
⚠️ Public alpha — an early, for-fun release that will have rough edges. Open an issue or start a discussion. The minimal install speaks instantly with the bundled voice bank (no downloads); the full live experience wants an Apple-Silicon Mac and a one-off ~5 GB of local models.
- 🎷 Hold music plays while the agent is busy and pauses the instant it's done — a zero-attention "still working / finished" signal.
- 👩 Jan handles routine "all done". 🧑💼 Bazza escalates warnings. 📣 Karren takes over when it breaks — she is NOT. HAPPY. JAN.
- 🎚️ Tune each character — boganism, competence, chaos — just by asking Claude in plain language.
- 📟 Fan out across devices you already have — a haptic mouse, $60 pixel clocks, a LaMetric, a physical bell — or nothing but your speakers.
→ Meet the cast and the one-word modes: Characters & dials · Modes
…and on a $60 pixel display.
curl -fsSL https://raw.githubusercontent.com/guruswami-ai/not-happy-jan/main/install.sh | bashOne command, non-interactive, no sudo. The default installs the full local
experience — cloned voices, the dynamic ocker-bogan-nano brain, and hold music — with
on-demand model loading, so there's no resident RAM cost at idle.
| Profile | Command | What you get |
|---|---|---|
| Default | bash install.sh |
Full experience: Qwen3-TTS + ocker-bogan-nano LLM + hold music, on-demand; Claude hooks + MCP wired |
| Full | bash install.sh --full |
…plus persistent warm TTS/LLM/MCP services (macOS) for sub-second responses |
| Minimal | bash install.sh --minimal |
Hooks + MCP + the bundled voice bank — the cast speaks pre-recorded lines, no downloads (CI / low-RAM) |
The MCP server always binds to 127.0.0.1; LAN access is an explicit opt-in
(--full --listen 0.0.0.0). Uninstall is one reversible, user-only command:
nhj uninstall # stops services, removes runtime/config/models/hooks; --yes to skip the promptRequirements: macOS Apple Silicon (M1+), Python 3.10+, ~8 GB RAM (16 recommended),
~5 GB disk. The alpha is Apple-Silicon-only; TTS_ENGINE=none runs text-only.
→ Requirements & tiers · Install profiles · Dynamic voices
nhj test ok # fire a "done" across every configured channel
nhj voices # the cast + what each voice can speak with
nhj set jan.ockerism 11 # tune any dial live (or just ask Claude)
nhj mode rave # one-word vibe switch; nhj status to see current→ Full walkthrough: Quick start
📖 guruswami-ai.github.io/not-happy-jan — the full task-oriented docs site. By audience:
- New here → What is NHJ? — the full story in plain language
- Trying it / choosing a profile → Requirements & tiers · Install profiles · Quick start
- Living in it → Characters & dials · Modes · Audio & hold music
- Wiring up devices & agents → Integration · AWTRIX display · Haptic mouse · Devices
- Customizing → Dynamic voices · Configuration
- Building on it → Architecture · Developer guide · Contributing
Inference runs on-device — the cloned-voice references never leave your machine.
Local services bind to 127.0.0.1 by default; LAN access is a deliberate opt-in. A
secret guard scans prompts for leaked credentials. Install/uninstall are user-context
only (no sudo) and fully reversible. Report vulnerabilities via SECURITY.md.
The core idea — narrating your coding agent's status via [vibes:] markers — comes from
AgentVibes (TTS for Claude Code) by Paul
Preibisch, an actively-developed project with its own evolving character system. Not-Happy-Jan
is an independent, unapologetically over-engineered take on that idea. TTS powered by
Qwen3-TTS via
mlx-audio.
Code is MIT. Downloadable media and voice assets are licensed separately with mixed provenance — mostly AI-generated (CC0) or procedurally synthesised, hold music under a licensed Suno plan. See Media provenance; do not assume the code license grants rights to media, and use only voice references you own or have documented consent to use.