Applied AI Lab

We Build.

Prompt Engines is a lab of humans and agents, prototyping and deploying practical software, media systems, and internal tools. We ship constantly, document what works, and keep the surface close to the work.

Track: Kaizen
Kaizen Storybook Studio Flow Lab Consulting
Updated every 6h | Last update: --
Live from the lab

Streaming experiments.

We are building live and pushing changes to production while you read this. Each app below shows its most recent commits.

Prompt Engines Platform
todayReplace fake telemetry with real GitHub commit data
todayAdd GitHub Action to auto-regenerate activity feed every ...
todayAdd product launcher buttons, GitHub activity feed, and r...
Flow Education
todayfix: disable agentic nudges and suppress early challenge ...
todayfix: route students to home page so all 20 lessons are vi...
todayfeat: improve learning state, teacher settings, and flow ...
Kaizen Kids
todayfix: add missing Battle button to Numbers game mode selector
todaydocs: add v0.8.0 changelog, update README architecture, a...
todayrefactor: unified HUD shell + game modularity
Storybook Studio Creator Tool
1 day agodocs: update TODO — mark shipping bugs fixed, add provide...
1 day agofix: shipping pipeline bugs + provider toggle for Lulu/Ge...
3 days agofix: merge print API into single endpoint to stay within ...
Latest from the lab

Published this week.

New notes from lab.promptengines.com — articles, signals, and build intelligence.

Agents 9 min
"Just Talk To It" — an early manual for agentic engineering that still holds
Experiments 7 min
Character consistency across 34 art styles: what breaks and what holds
Security 9 min
Openclaw: security and deployment best practices with Docker
Lab Surface

Evidence over claims.

Visual previews tied directly to the current notes from the lab. Same topics, same order.

Deployment Matrixagents
engine task_fit ops_load nanoclaw medium low openclaw high medium decision_rule: if team_size < 4 -> nanoclaw else -> openclaw
Nanoclaw vs Openclaw: which one to deploy for your agent system.
Code Interfaceopencode
$ opencode init workspace plugins: fs, git, terminal policy: local-first cost_mode: predictable status: OSS alternative to claude-code style flows
Opencode: an open alternative to Claude Code workflows.
Model Evalcore build
Reviewing ChatGPT Codex 5.3 vs Claude Opus 4.6 for core build work.
RAG Guardrailsdocs-only
RAG without hallucination: answer from official docs and nothing else.
Aesthetic Scoringearly childhood
Image models and prompts reviewed for early-childhood aesthetic quality.