We open-source the hard parts.
Three of our engineering bets, in public. Same people, same standards as the work we ship for clients, only with the source open and the tests visible. It's the same engineering we bring to your work.
Nebula, private AI for your knowledge, running entirely in your browser.
A local-first knowledge tool that runs retrieval-augmented generation over a real entity graph (GraphRAG), with on-device embeddings and an in-browser LLM. No server, no account, no install, nothing leaves the tab.
Privacy-preserving AI · local-first / on-device LLM · WebGPU + WASM · GraphRAG · zero-install
Kite, an agent framework that treats the LLM as untrusted.
Production scaffolding for AI agents: the LLM proposes, a kernel you control validates, with circuit breaker, kill switch, idempotency and 5 reasoning patterns (ReAct, ReWOO, Tree-of-Thoughts, Plan-Execute, Reflective). The difference between a demo that works once and an agent you can put in front of real users.
AI agents · agentic systems · LLM safety & guardrails · production reliability · open-source infra
Cosmos, the observable universe in your browser, from real astronomical data.
Browser-native 3D rendering of real astronomical data (Gaia DR3, JPL SPICE) with procedural GLSL and a Rust tile server over PostGIS and Elasticsearch. Our proof for large-scale data and real-time visualization. Every pixel is math.
Large-scale data · real-time 3D · GLSL · Rust · PostGIS + Elasticsearch · open-source