Mellow Machines Rules react immediately.
Taps, task changes and attention signals can trigger feedback as soon as the event arrives.
Mellow Machines Lab
Instant Response.
Infinite Character.
Mellow Harness is built around System One Models like Jev to empower AI hardware and applications with fast, expressive feedback.
PS: It is 100% open source.
Try Mellow Harness in Action.
Scripted · 4 min
events.jsonl
You · speaking…
Waking the buddy up…
StateStartingMoodCalm
0:00 / 4:00
A coding day, with a little company.
Scripted tasks, taps and conversations. User speech is shown as text. Enable sound for Boop’s recorded voice. No live model calls; timing is illustrative.
The harness behind the character
Quick feedback. Room for a little judgment.
Immediate reactions come from code. A System One Model chooses what fits the moment. One shared event log keeps both paths in context.
The model chooses.
Personality, recent events and an allowed action roster guide Jev’s next decision.
Both paths leave a trace.
Code and model outcomes append to the same log, which feeds the next iteration.
In the Buddygotchi example
Character at three timescales.
Personality
Authored character in the prompt. Nightly learning is work in progress.
Mood
The model moves through related moods, with context for each transition.
State
The current task becomes visible: working, waiting, or finished.
Build with the Lab
Start with the piece you need.
01 / Agent harness
MellowHarness
The loop, bounded model choices and shared event log for responsive agents. Bring your own character, actions and outputs.
Start without a model key.
A scripted brain exercises the same runtime.
02 / Companion example
Buddygotchi
A little companion that follows your coding agents. Animations, sounds and connected moods make the coding day feel a little more alive.
Bring it to your desk.
The current device build uses a Mac host and an ESP32 display.
03 / Agent integration
AgentHooks
Claude Code and Codex hooks, normalized into one event stream. Give your application work and attention signals from coding agents.
Use it beyond the companion.
A small integration layer for applications that follow coding agents.
Lab profile and community files: .github repository.
Technical sharings & build guides
From the loop to a face on your desk.
Start with harness design and decision models, connect a coding companion, then follow the animation and ESP32 build. Hardware is one way to bring the idea to life.
The public hardware build
Build your own Buddygotchi.
Agent events → mood choices → pixel animation → ESP32 display. Our walkthrough covers the current public Pixel pack and supported CYD board.
Architecture guide
Browse all technical sharings and resources →
Agent harness design for fast, expressive feedback
How MellowHarness combines a System One decision loop, bounded action choices, immediate rules and a shared JSONL event log for AI hardware and applications.
Decision model guideSystem One Models: Jev and typed decision interfaces
A source-backed guide to Jev, OpenAI Decisions API and the open-source Laya project. Typed action choices, integration status and evaluation boundaries.
Measured application sampleJev decision-pass latency: a 105-pass application log
An audited historical Jev sample: 227 ms median across 105 decision passes. Full distribution, sanitized data and limits of the measurement.
Quickstart · No model key or hardwareTry MellowHarness without a model API key
Run Beacon, send events and inspect how rules and model choices meet in the log.
Companion architectureBuild an expressive companion for Claude Code and Codex
Connect coding-agent hooks to Buddygotchi’s state, mood and character. MellowHarness architecture, public build requirements and scripted-demo boundaries.
Build guide · Harness → Animation → HardwareBuild Buddygotchi: from harness to animation to hardware
Follow the public Pixel build: coding-agent events, connected moods, animation packs, a CYD display and the first device checks.
Why Mellow Machines?
Make room for something warm.
Mellow Machines Lab is a small open-source team applying frontier harness principles to build warm, welcoming, playful companions.
We feel the noise and anxiety around AI too. This is our after-work space to make something fun and heartwarming—and share the tools that help other builders do the same.