Discover the Best AI Resources

vllm-omni

Extends vLLM beyond text to serve omni-modal models — Qwen3-Omni, TTS like CosyVoice3, and diffusion image/video/audio generators — in one engine, adding the non-autoregressive Diffusion Transformer support the core project never targeted.

Gives coding agents controlled access to a live Chrome browser through MCP, exposing DevTools-backed automation, debugging, screenshots, console inspection, network analysis, and tracing.

Human Psychometric Questionnaires Mischaracterize LLM Behavior

Compares standard human psychometric questionnaires (PVQ, BFI) with generation‑based profiling to test whether questionnaires predict real LLM responses. Finds big divergences: questionnaires exploit lexical cues and elicit alignment‑consistent answers, mischaracterizing LLM behavior on everyday queries.

Eight example apps for building with the Claude Agent SDK: an IMAP email assistant, a multi-agent research system, an Excel agent, a React/WebSocket chat UI, a .docx resume generator, and hello-world session demos. Local-only, not production.

APM – Agent Package Manager

Declares and installs agent dependencies from an apm.yml manifest—skills, prompts, agents, plugins and MCP servers—with transitive resolution, security auditing, plugin packaging, and cross-host registries so agents are reproducible across repos.

DeepAudit

Audits source code for security flaws using LLM agents, then auto-generates and runs proof-of-concept exploits in Docker sandboxes to confirm which findings are real. Retrieves CWE/CVE knowledge via RAG; runs on hosted or local Ollama models.

Automatically removes safety alignment from transformer LLMs via directional ablation, with Optuna's TPE optimizer tuning the parameters — no retraining or model-internals expertise needed; hit 3/100 refusals at 0.16 KL on Gemma-3-12b.

FluidVoice

On-device macOS dictation that transcribes speech locally and offers an optional local AI enhancement (Fluid Intelligence) for smart formatting and post-processing. Key features: low-latency model choices, live transcription overlay, per-app prompts and privacy-by-default; best on Apple Silicon.

LongLive 2.0: An NVFP4 Parallel Infrastructure for Long Video Generation

Provides an NVFP4‑optimized training and inference infrastructure for long-form video diffusion models — supports multi-shot AR training, KV-cache and NVFP4 quantized inference, sequence-parallelism and async decoding for higher FPS and longer outputs.

Anthropic / skills

Folders pairing a SKILL.md (YAML metadata plus instructions) with optional scripts that Claude loads on demand to handle specific tasks. Bundles document skills (PDF, DOCX, PPTX, XLSX), a spec, and a starter template for authoring your own.

Embeds a GUI agent in your web page as client-side JavaScript, letting users drive the interface with natural language — it reads the DOM as text (no screenshots) and performs clicks and form fills. Bring your own LLM; no extension or backend required.

A2UI: Agent-to-User Interface

Lets AI agents describe interactive UIs as declarative JSON instead of executable code; client apps render the components with native widgets from a pre-approved catalog, keeping agent-generated UI safe across trust boundaries.