Search

Automatic cross-provider model fallback with quota awareness for the DeepSeek Harness

PGC business domains — the capability implementations the composed reference snapshot binds at execution

A simple Django app for sending template-driven emails.

QArmor blockchain side: the DVN worker and QArmorDVN.sol. Producer zero of the telemetry pipeline (INV-10).

n8n community node for UnRemind.me — context-aware reminders that surface when your situation matches, not at a fixed time

Local OpenAI twin — a faithful, stateful local OpenAI API your real `openai` SDK talks to unmodified. The model is stubbed (deterministic) and embeddings are deterministic pseudo-vectors, but the protocol envelope (chat completions/streaming/tool_calls, t

A/B test harness for LLM prompt and skill changes — headless Claude Code or any OpenAI-compatible endpoint.

Pluggable pipeline engine for splitting work into sequential steps backed by interchangeable stores

binary computing engine that turns distributed storage into a publishable working set carrying the merged virtual-hardware engine (the saddle + e2ugh grand merge)

NexusCompute Unified SDK — renter, provider, and local Nexus Agent

Zero-runtime multi-agent orchestration convention for Claude Code, OpenCode, Codex, Antigravity, and Cursor

Local kit for ingesting corpora, synthesizing LLM eval questions, reviewing gold, and judging model answers.

Use deck.gl as a custom Google Maps overlay

Production-grade Discord MFA authentication library. Connection pooling, TOTP, Cloudflare bypass, automatic token refresh. Zero runtime dependencies.

Official portable agent skills for Mostly Good Metrics

CDFI and MDI peer benchmarking tool using FDIC call report data — NIM, efficiency ratio, ROAA, ROAE, Tier 1 leverage, and more

Prebuilt codeanalyzer-iac (caniac) backend binary for CLDK (codellm-devkit).

AI-powered subtitle translator for SRT files, supporting OpenAI and Gemini.

Trim a plain OpenAI-shaped message array down to a token budget without ever orphaning a tool result.

Reorder LLM prompt blocks from least- to most-volatile so provider prefix caches stay hot, and measure whether it actually worked.