Tag

DFlash: Block Diffusion for Flash Speculative Decoding

Enables parallel speculative decoding by using a lightweight block-diffusion draft model to produce multi-token drafts for faster, high-quality generation. Integrates with vLLM, SGLang and Transformers backends and ships draft models on Hugging Face.

notebooklm-py

Provides a Python API, CLI, and agent skill to programmatically access Google NotebookLM — exposing features the web UI omits (batch imports/exports, PPTX slides, mind‑map JSON) and integrating with LLM agents like Claude Code and Codex.

PersonaPlex

Real‑time full‑duplex speech‑to‑speech system that controls conversational role via text prompts and voice timbre via audio-conditioned embeddings. Built on Moshi; optimized for low-latency, persona-consistent spoken interactions.

Headroom

Compresses any context sent to LLMs (tool outputs, DB reads, RAG results, files, logs) to cut tokens by ~70–95% while preserving reversible originals; runs as a proxy or Python/TypeScript SDK with integrations for common agent frameworks.

World Monitor

Aggregates global news, infrastructure, military and market signals into an interactive map dashboard and synthesizes AI-generated intelligence briefs. Key features: local/remote LLM support, 3D globe + flat map, 35+ data layers, country instability index and client-side RAG/embeddings.

daily_stock_analysis

Generates daily LLM-powered decision dashboards for A/H/US stocks by combining multi-source market data, real-time news, technical signals and agent-style strategy reasoning; deploys via GitHub Actions or Docker and pushes reports to multiple channels.

Stop Slop

Teaches LLMs to detect and remove “AI tells” from prose using curated phrase/structure lists, before/after examples, and a 5‑dimension scoring rubric. Delivered as a reusable skill (SKILL.md + reference files) designed to plug into Claude or any LLM workflow for automated style sanitization.

Engram — Conditional Memory via Scalable Lookup

Provides a conditional memory module that performs O(1) N‑gram lookups and fuses static embeddings into transformer hidden states — enables offloading large embedding tables to host memory with minimal inference overhead.

GenericAgent

Grows a personal skill tree by crystallizing each solved task into reusable skills; a ~3K-line autonomous agent framework that gives an LLM system-level control of browser, terminal, filesystem, input/vision and mobile (ADB) via nine atomic tools, optimized for low token cost.

CodeWhale

Terminal-native coding agent that streams reasoning blocks, makes controlled edits to local workspaces behind approval gates, and includes an auto mode that chooses model and thinking level per turn — designed for in-terminal code review, debugging, and automation workflows.

Craft Agents

Desktop-first agent client that composes LLM-driven agents into document-centric, multi-session workflows; it wires APIs, MCPs and local tools into shareable sessions, supports multiple LLM providers, and exposes a headless server + CLI for automation.

Craft Agents

Desktop + CLI agent-native client for managing multi-session conversations, connecting to multiple LLM providers and external data sources, and creating shareable agent skills and automations without editing code.