Browse agent skills

Profile data into human-readable copy and UI chrome using heuristics — better UX from the rows you already have.

Embeddable React AI Chatbot & Customer Support Widget powered by AIHub360. Plug-and-play customizable floating chat assistant with website FAQ training, TypeScript definitions, and Next.js / Vite support.

Automatic, cost-tiered vision model fallback plugin for OpenCode — when an agent gets an image, it's automatically routed to a model that can actually see it.

Self-hosted cockpit for running a fleet of AI coding agents across your own machines — parallel worktree runs, live board, compare & merge, web terminal, design capture.

Framework-neutral form schemas, evaluation, validation, scoring, localization, and automation planning.

App-agnostic bounded agent loop and streaming assistant interface, with provider policy supplied by the host app.

Avatar parts library, composition model, and the composer — one full-body figure per subject, portraits derived by head viewport; plus AI image generation through the AI SDK provider layer.

A pure UI and state package for a chat widget — zero runtime dependencies, no network calls. Bring your own onSendMessage.

SodaMem memory layer for the DeepSeek Harness (Cordis): unconditional per-turn recall and per-turn-close retain.

Photon-hosted iMessage transport and configuration UI for DeepSeek Harness

BDB Synapse — 3D Codebase Visualizer & Agent Session Replay Engine

A modern Python client for the DomainIQ API - Domain Intelligence and Security Research

The distributable self-hosted workflow runner for Jentrix.

Trajectory For All — a browser-based inspection surface for pi coding agent sessions.

Interactive CLI assistant for building and running docker run commands

This package is built from the AI SDK fork at [`zshannon/ai@6c1159713`](https://github.com/zshannon/ai/commit/6c1159713e200cfaa37b5b1b2a1d0d94a5971ba3). It preserves OpenTelemetry model-attempt spans across `streamText` retries.

OpenAI-compatible AI SDK providers for Meta (muse-spark), Alibaba DashScope (glm-5.2, kimi-k2.7-code, qwen3.8-max) and Amazon Nova (nova-2-lite-v1) with tool calling support.

Bind a voice agent to your web app — compile your codebase into an agent-readable capability manifest, then talk to your app.

OpenAI-compatible AI SDK providers for Meta (muse-spark), Alibaba DashScope (glm-5.2, kimi-k2.7-code, qwen3.8-max) and Amazon Nova (nova-2-lite-v1) with tool calling support.

OpenAI-compatible AI SDK providers for Meta (muse-spark), Alibaba DashScope (glm-5.2, kimi-k2.7-code, qwen3.8-max) and Amazon Nova (nova-2-lite-v1) with tool calling support.

Modular search tool for DeepSeek Harness: attach SearXNG, RAG and your own sections to the native web_search.

Agent skills (SKILL.md) that find and fix the five security failures AI-generated code leaves behind most often: missing RLS, frontend-only authorization, IDOR, exposed secrets, unvalidated input. Works in Claude Code, Codex, OpenCode, Antigravity, Devin,

OpenAI-compatible AI SDK providers for Meta (muse-spark), Alibaba DashScope (glm-5.2, kimi-k2.7-code, qwen3.8-max) and Amazon Nova (nova-2-lite-v1) with tool calling support.

Benchmark framework: task + step + config primitives and the local orchestrator that drives @benchsdk/worker

Offline text-to-speech for React Native, powered by Piper

Data and server layer for building Whop templates on the Whop REST API

Engelbart: one-command installer for /goals-ui, chat-scoped goals for Claude Code.

Convert anything to markdown. PDF, DOCX, PPTX, XLSX, HTML, EPUB, Jupyter, RSS, images, audio, URLs, and more. Works as a CLI and as a library.

Pi extensions for the web: fetch_markdown (URL to markdown, cached, keyless), pdf_extract (PDF text extraction) and web_browser (headless Playwright automation), sharing a preapproved-host allowlist.

stratusd — the always-on Stratus Agent gateway: durable SQLite sessions, a per-provider runner pool, dispatcher-backed delegation, and an activity watchdog