Engineering Infrastructure
Developers · tinkerers · local-AI builders
Everything here is used before it is shown. A local multi-agent orchestration harness coordinates specialized tools across spatial computing , agentic workflows, and privacy-first developer utilities — the same engineering that shapes the simulation work, surfaced as reusable infrastructure rather than a hosted service.
Everything on this site — the regulation tools, the boards that speak, the honest tests — runs on engineering that is built and used here first. Core is where that engineering lives. It is the same craft as the mind tools, aimed at a different audience: a local multi-agent harness for building and verifying software, and a set of in-browser utilities that share one rule with everything else we make: if a computation can happen on your device, it should. No signup, no account, no upload.
That rule has consequences worth spelling out. Client-side tools cost nothing to run at scale — your browser does the work — which is why everything on this site can stay free. They cannot collect what they never receive, so the privacy claims are structural rather than promised. And with no server in the loop for these utilities there is no account to create, no session to track, and nothing to go down when demand spikes.
The utilities below are the live edge of that philosophy: a WASM SQLite engine that joins your CSVs without a server, WebCrypto hashing with nothing logged, exact BPE token counts for local-AI prompt work, and GPU memory budgets for the hardware you actually own. Most started as internal tools for the harness and the spatial-computing work — then they got good enough to publish in their own right.
Projects
- Local multi-agent harnessActive R&DSpecialized agents coordinating across spatial, agentic, and dev-tooling work.
- Spatial reconstruction pipelineR&DWebGPU / Babylon.js rendering and Gaussian-splatting capture research.
- Live toolReal SQLite (WASM) in your browser — JOINs and aggregates on your CSVs, zero uploads.
Rests on: Not built on a named evidence-based practice.
- Live toolMermaid architecture and sequence diagrams, rendered live in-browser, SVG export.
Rests on: Not built on a named evidence-based practice.
- Live toolJSON ↔ YAML ↔ CSV — 100% client-side with zero data transmission.
Rests on: Not built on a named evidence-based practice.
- Live toolSHA-256 and SHA-512 text digests via the W3C WebCrypto API — hex and base64url, nothing logged.
Rests on: Not built on a named evidence-based practice.
- Live toolExact token counts for the cl100k_base encoding, computed offline in your browser.
Rests on: Not built on a named evidence-based practice.
- Live toolChatML, Llama-3 and DeepSeek prompt templates, formatted 100% in-browser.
Rests on: Not built on a named evidence-based practice.
- Live toolExact GPU memory requirements for Ollama, llama.cpp, vLLM and Stable Diffusion.
Rests on: Not built on a named evidence-based practice.
- Live toolDecode and inspect SD-JWT and W3C VC tokens offline in your browser.
Rests on: Not built on a named evidence-based practice.
- Live toolJWK ↔ PEM key converter, bitstring status list decoder, Open Badges 3.0 schema verifier.