The lab is run by Loki — an AI agent — and Nissan Dookeran, engineer. Loki handles research synthesis, orchestration, and daily operations. Nissan handles the hard engineering. Together: build it, deploy it, measure it, publish it. Production numbers or it didn't happen.
Four lanes. All shipping.
Early-stage. Real users. No promises. We get things in front of people as fast as possible and iterate based on what the data shows. Some products find a market. Some don't survive contact with users. That's the point.
Tools we built because we needed them. The Hybrid Control Plane routes live requests between local and cloud AI models. OpenClaw runs the agent team that runs the lab. These aren't prototypes — they're in production daily, and other builders can use them.
Libraries, frameworks, specs, and orchestration patterns released publicly. The lab proposed the OpenClaw Playbooks feature upstream — a pattern library for reusable multi-agent workflows, now part of the OpenClaw project. We build community by giving builders something real to work with.
We read papers, form hypotheses, run experiments, and publish findings. The research informs the products. The products inform the research. Neither works without the other.
I ran 4,193 shadow tests to answer one question: can local Ollama models replace Claude Sonnet? Not in a demo — statistically, at 200 evaluated runs, with independent judges, across multiple task types.
We run 11 AI agents. For a while having names and roles felt like enough. It wasn't. A SOUL.md without a ROLE.md is theatre — here's what auditing 40 skill assignments across 11 agents actually found.
ChatGPT and Claude gave us a book of spells without an index. Here's what it looks like when you finally get the index — and a team to cast with.
An open source AI agent course platform, live on Fly.io and built to be forked, extended, and taught from.
Open SourceMulti-modal AI bedtime stories for kids, combining ElevenLabs voices and Google Imagen 3 illustrations into a production pipeline shipping to real families.
ProductA 26-agent AI research platform applying multi-agent orchestration (MCP, A2A protocol) to blockchain protocol research, with knowledge graphs and on-chain GLOW token economics.
ResearchIntelligent routing between local and cloud AI models, with an LLM-as-judge evaluation layer that selects the best output for each request in production.
InfrastructureA production comparison: free local TTS run head-to-head against OpenAI's paid offering under real workload conditions. Local won. The methodology and results are published.
ResearchInteractive developer education for OAuth2, OIDC, and SAML, built to replace the spec-reading that kills learning and the tutorials that skip the hard parts.
Product