Sam Wasserman

eyes-of-kai. A local-first AI companion that keeps watching while nobody is talking, so it can tell you what you were doing five minutes ago. Runs entirely on your own hardware.

circle-take. The dailies room for AI filmmaking — organize every generated take by shot, gate them with local AI quality checks (character drift, flicker, anatomy, continuity), compare, circle the winners, and export selects + reshoot lists with exact prompts and seeds. No API keys. Apache-2.0.

GLM-5.2-Abliterated-NVFP4-316K-4x-DGX-Spark. Abliterated GLM-5.2 with a true 4-bit NVFP4 KV cache on 4x DGX Spark: 316K context, 317,279-token pool (+58.6% vs fp8), 41.4 tok/s peak — with honest content-dependent decode measurements.

GLM-5.2-NVFP4-AQLM-Triple-DGX-Sparks. GLM-5.2 NVFP4+AQLM on 3× DGX Spark — 380k context MTP serve stack

no-ai-slop. Removes 20+ patterns of AI slop from any piece of writing.

autonomous-vibe. Let's vibe hardware: create magical things by chatting with AI.

ComfyUI-SeedVR2_VideoUpscaler. Official SeedVR2 Video Upscaler for ComfyUI

FlashVSR. [CVPR 2026] Towards Real-Time Diffusion-Based Streaming Video Super-Resolution — An efficient one-step diffusion framework for streaming VSR with locality-constrained sparse attention and a tiny conditional decoder.

Laguna-S-2.1-DGX-Spark-RTX-6000-PRO. vLLM 0.25.1 serving stack for poolside/Laguna-S-2.1-NVFP4 with DFlash speculative decoding — DGX Spark & RTX 6000 PRO

codex-build. Orchestrator drives, Codex codes — execute an approved plan one task at a time with a test gate before every commit and exactly one PR at the end

GLM-5.2-QuantTrio-200K-4x-DGX-Spark--36tok-s. Recipe: GLM-5.2 (unpruned QuantTrio Int4-Int8Mix) at 200K ctx with MTP spec decode on a 4x NVIDIA DGX Spark (GB10) cluster

dgx-spark-2x-deepseek-v4-flash. Reproducible kit to deploy DeepSeek-V4-Flash-DSpark on a 2× NVIDIA DGX Spark (GB10) cluster: vLLM TP=2 over QSFP 200GbE, NVFP4 KV, DSpark speculative decoding, 1M context, systemd self-heal. Apache-2.0.

Minimax-M3-NVFP-3x-DGX-Sparks-TP-3. Working recipe: MiniMax-M3 NVFP4 at tensor-parallel 3 across 3x DGX Spark (GB10/sm_121) with clean tool-calling. Includes the head-node OOM fixes and multi-node Ray/NCCL setup. Open for tinkering + fixes.

cork-board. The digital cork board for filmmakers. Index cards, acts, arcs, and episodes — a local-first planning wall for features, TV, shorts, music videos, commercials, and docs. macOS, Windows & Linux.

scriptbreak. Screenplay breakdown & AI prompt-pack studio — import a script, get scenes, elements, bibles, shot lists & a timeline, then export self-executing prompt packs for any LLM and any video/image generator. No API keys.

vLLM-Moet. DGX Spark (GB10/sm_121) port: DeepSeek-V4-Flash 159B on one Spark, GLM-5.2 753B across two. Prepacked 2-bit planes + unified-memory fixes on top of upstream vLLM-Moet — see spark/RUNBOOK.md

tool-eval-bench. Tool-calling quality benchmark for LLM serving stacks. 80+ deterministic scenarios testing multi-turn orchestration, safety boundaries, and structured output. Supports vLLM, SGLang, and llama.cpp.

hermes-davinci-resolve-plugin. Hermes Agent plugin for DaVinci Resolve Studio live control and free Resolve interchange workflows

master-canvas-hermes-plugin. Hermes Agent plugin and bundled skill for Master Canvas AI video handoff packages.

master-canvas-mcp. Headless MCP server for Master Canvas — read/edit a project (boards, cards, prompts, shot order, assets) and build generator-ready handoff packages from Hermes, Claude Code, Codex, or any MCP client. Zero dependencies. MIT.

sam-pdf-studio-mcp. Headless MCP server for Sam PDF Studio — run its PyMuPDF engine (merge, split, redact, OCR, convert, page ops) on local PDFs from Hermes, Claude Code, Codex, or any MCP client. Zero dependencies. Apache-2.0.

sam-pdf-studio. A native macOS PDF editor — edit text in place with matching fonts, annotate, fill & sign with saved signatures, redline, OCR, and manage pages. SwiftUI/PDFKit over a local PyMuPDF engine. Apache-2.0.

freecut. Fork of browser-use/video-use with the paid ElevenLabs dependency replaced by free, pluggable transcription (local Whisper by default; VibeVoice-ASR for diarization).

blockout. Open Source Desktop App to help with Previs for AI-native filmmaking — stage grey-box scenes, choreograph camera & cast with marks, export motion-reference packages for Seedance/Veo/Kling/LTX/Wan. Apache-2.0.

hermes-world-landing. Public landing page for hermes-world.ai — premium dark-fantasy Agentic MMO marketing site.

clipper-cowboy. A local-first shot library for AI video IPs. Catalog characters, scenes, and ready-to-use clips.

wassermans-filmmaker-suite. Wasserman's Filmmaker Suite — ScriptBreak, Cork Board, Master Canvas, Blockout, Motion Previs Studio, Storyboard Reference Studio, Circle Take, Stem Studio, and the Unofficial DaVinci MCP: AI-native filmmaking from script to final mix.

DeepSeek-v4-Flash-DSpark-Abliterated-Uncensored-2x-DGX-Spark. Keys/drowzeys' abliterated (uncensored) DeepSeek-V4-Flash-DSpark served on 2x DGX Spark TP=2. Fork of the DS4 DSpark recipe with the model swapped in. Full model credit: Keys/drowzeys.

ZohoVideoRenamer. Python

unofficial-davinci-mcp. Let any AI agent run DaVinci Resolve — MCP server with live Studio control, free-edition FCPXML/LUT workflows, reference color matching, beat-aware music cuts, dialogue tightening, broadcast-loudness mixes, editorial knowledge, and a local push-to-talk voice bridge. Apache-2.0.

git. "the information manager from hell" - linus torvalds [e83c516, 7 Apr 2005]

motion-previs-mcp. MCP server for Motion Previs Studio — extract camera + performance motion from reference footage via any MCP agent (Hermes, Claude, Codex). Zero dependencies.

stem-studio. Split any married mix into Dialogue, Music & SFX — local AI stem separation for filmmakers. macOS & Linux.

2Wild-Vision-Mode. Give any blind LLM eyes — on any machine — in one shot. Tiny local VLM (Qwen3.5-0.8B) as an OpenAI-compatible vision service. Mac (MLX) / PC+NVIDIA (llama.cpp) / DGX Spark.

blockout-mcp. MCP server for Blockout — drive the open-source previs app from any MCP agent (Hermes, Claude, Codex). Zero dependencies.

storyboard-reference-studio. Turn any reference imagery into storyboard stills + image-generator prompts. Part of the Blockout previs suite. Apache-2.0.

GLM-5.2-NF3-Hybrid-4x-DGX-Spark-800kctx. Python

keys-vLLm-0.24.0-Optimized-DeepSeekV4-Flash-DSpark-NVFP4-KV-1.5M-CTX-3M-Pool-C-12-on-2-DGX-Spark. Run on TWO-DGX-Spark - vLLm-0.24.0 dual cache optimized DSV4F+DSpark+NVFP4 KV (Concurrency 12 with 1.5M context/3M KV token Pool) >0.58-0.6 acceptance.

motion-previs-studio. Open-source desktop app for AI-film motion, depth, pose, and camera-move previsualization.

Hy3-295B-NVFP4-MTP-2x-DGX-Spark. Tencent Hunyuan 3 (295B MoE) on 2x NVIDIA DGX Spark: NVFP4 W4A16 + native MTP speculative decoding. First published MTP-on-GB10 numbers, full recipe, KV math, and every bring-up bug documented.

dgx-vllm. Shell

alexa-skill-llm-intent. Alexa Skill that provides turn based conversations with an AI LLM. Bringing AI to your Alexa, because Amazon doesn't.

Qwen3.6-35B-A3B-UD-Q8_K_XL_DGX-Spark-Recipe. llama-server start/stop scripts for Qwen3.6-35B-A3B UD-Q8_K_XL GGUF on DGX Spark

MiMo-V2.5-DFlash-1M-ctx-NVFP4-KV-2x-DGX-Spark. MiMo-V2.5 NVFP4 4-bit weights + NVFP4 4-bit KV cache + DFlash speculative decoding on 2× NVIDIA DGX Spark — 1M context, 3.4M-token KV pool

MiniMax-M3-2x-DGX-Spark-36-tok-s. MiniMax-M3 (428B, no pruning) at 36 tok/s on 2× NVIDIA DGX Spark — W4A16 GPTQ + NVFP4 KV + EAGLE-3 speculative decoding on vLLM. Three serving lanes: speed / balanced / long-context.

MiMo-V2.5-NVFP4-2x-DGX-Sparks-TP-2. Tuned recipe: Xiaomi MiMo V2.5 (NVFP4) on 2x DGX Spark, vLLM TP=2 over RoCE. ~32-33 tok/s, Quality 89.9, 160K context, omnimodal + tool-calling. Non-eager + MTP=2.

Keys---Full-GLM-5.2-Quantrio-INT4-INT8-mixed-8bit-Attention-on-4-x-DGX-Spark-GB10-Cluster. Updated (Latest): Full (non-pruned) GLM-5.2 DSA serving TP=4 on 4×DGX-Spark/GB10 via rebuilt sm_121a vLLM — Updated now with NVFP4 KV + increased context to 100ktarter pack for 4 DGX-Spark (GB10) Cluster over CX-7 Fabric running full GLM 5.2 locally (Quantrio INT4-INT8 mixed) 8Bit attension This model is non-prune non-REAP no lobotomy.

MiMo-V2.5-TP3-NVFP4-KV-3xDGX-Spark. MiMo-V2.5 Omni · TP=3 on 3x DGX Spark · NVFP4 4-bit KV cache → ~3.5x more 1M-context KV pool (3.06M→10.59M tokens) vs FP8

MiMo-V2.5-TP2-1M-NVFP4-KV-2xDGX-Spark. MiMo-V2.5 Omni TP=2 on 2x DGX Spark · 1M context · NVFP4 4-bit KV (~1.97M-token KV pool @ 1M, ~30 tok/s) · 69-eval: thinking-OFF 97.8 beats thinking-ON 90.6 for tool/agent work

MiniMax-M3-AWQ-1M-NVFP4-KV-4x-DGX-Spark. MiniMax-M3-AWQ serving 1,048,576-token (1M) context on 4x DGX Spark (GB10) via a 4-bit nvfp4 KV cache. Built on CosmicRaisins' M3 recipe.

DeepSeek-v4-Flash-DSpark-2x-DGX-Spark. Python

Keys-Concurrency-Patch-for-DSpark-DeepSeek-V4-Flash. Concurrency Patch to enable on DSpark - DeepSeek V4 Flash/Pro C=16 can be modified for more

DeepSeek-v4-Flash-DSpark-1M-NVFP4-KV-2x-DGX-Spark. DeepSeek V4 Flash DSpark 1M NVFP4 KV recipe for 2x DGX Spark

DeepSeek-v4-Flash-DSpark-60-tok-s-900K-ctx-2x-DGX-Spark. Self-contained DeepSeek V4 Flash DSpark TP=2 recipe for 2x DGX Spark with 62 tok/s benchmark

DeepSeek-v4-Flash-DSpark-2x-DGX-Spark. DeepSeek-v4-Flash recipe for 2x DGX Sparks

DeepSeek-V4-Flash-Dual-DGX-Spark-1M-Context. Deploy DeepSeek V4 Flash (MoE reasoning model) on dual DGX Spark nodes with 1M token context, InfiniBand, and FP8 KV-cache

mercury-agent. Soul-driven AI agent with permission-hardened tools, token budgets, and multi-channel access. Runs 24/7 from CLI or Telegram.

Rebalance-Pack. A set of evolving nodes that streamline the best use practices in ComfyUI.

ComfyUI-Majoor-AssetsManager. Majoor Assets Manager is an advanced asset browser for ComfyUI that provides a comprehensive solution for managing, organizing, and viewing your generated assets. It integrates directly into ComfyUI, offering features like full-text search, metadata extraction, rating and tagging systems, and advanced viewing capabilities.

Inline-Studio. AI filmmaking on a node canvas. Build your whole visual pipeline from moodboard to final cut. Generate on your own GPU with diffuser based generation engine. Train your own Z-Image & Krea2 lora models.

krea-2. Official inference code for Krea 2

Qwen3.6-27B-AEON-Ultimate-Uncensored-DFlash. Fully uncensored, capability-enhanced abliteration of Qwen3.6-27B. NVFP4 + z-lab DFlash speculative decoding (n=12) on the unified ghcr.io/aeon-7/aeon-vllm-ultimate:latest container, tuned for long-context draft acceptance on DGX Spark. 6 HF variants (BF16/NVFP4/MTP/MTP-XS), docker-compose, and QuickStart.

oh-my-pi. ⌥ AI Coding agent for the terminal — hash-anchored edits, optimized tool harness, LSP, Python, browser, subagents, and more

Gemini-Watermark-Remover. An AI powered extension to get rid of the Gemini watermark

hyperframes. Write HTML. Render video. Built for agents.

vllm-ultimate-dgx-spark. AEON vLLM Ultimate — vLLM 0.25.0 built from source for DGX Spark / Blackwell (sm_121a/GB10). One image serves the whole AEON fleet (Gemma-4-26B-A4B, Qwen3.6-27B, Qwen3.6-35B-A3B) with DFlash spec-decode on a pinned V1 runner, Triton NVFP4-KV, FP8 KV, NVFP4 swizzled-scale decode fix, FlashInfer 0.6.13, TP=2-ready.

hyperframes-launches. Open-source HyperFrames compositions behind HeyGen's product launch videos.

cosmos. NVIDIA Cosmos is an open platform of world models, datasets, and tools that enables developers to build Physical AI for robots, autonomous vehicles, smart infrastructure, and more.

Agent-Reach. Give your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu — one CLI, zero API fees.

ai-data-extraction. extract all your personal data history from cursor, codex, claude-code, windsurf, and trae

skills. Agent Skills for Google products and technologies

supervision. We write your reusable computer vision tools. 💜

cosmos-locateanything-dgx. DGX Spark pipeline: Cosmos 3 generated video plus Locate Anything live grounding demo

NemoClaw. Run agents like Hermes, LangChain Deep Agents, and OpenClaw more securely inside NVIDIA OpenShell with managed inference

apple-notes-exporter. MacOS app written in Swift that bulk exports Apple Notes (including iCloud Notes) to a multitude of formats preserving note folder structure.

obsidian-importer. Convert your data to Markdown files you can use in Obsidian. Works with Apple Notes, OneNote, Evernote, Notion, Google Keep, and many other formats.

video-search-and-summarization. NVIDIA AI Blueprint for video search and summarization (VSS) is a GPU-accelerated reference architecture for building video analytics agents with real-time verified alerts, visual Q&A, and automated reporting. The VSS Blueprint uses vision language models (VLMs) such as NVIDIA Cosmos, LLMs such as NVIDIA Nemotron, RAG, and NVIDIA NIMs.

WhatDreamsCost-ComfyUI. LTX Director and a variety of other custom ComfyUI nodes and workflows

master-canvas. Open Source Desktop App for Local-first pre-production canvas for AI video planning, prompts, assets, and handoff packages.

awesome-seedance-2-prompts. 🎬 2000+ curated Seedance 2.0 video generation prompts — cinematic, anime, UGC, ads, meme styles. Includes Seedance API guides, character consistency tips, and advanced video workflows.

designmd. Production-grade design context for AI coding workflows. Extract a real design system from any URL — colors, typography, spacing, breakpoints — as a portable DESIGN.md.

ViMax. "ViMax: Agentic Video Generation (Director, Screenwriter, Producer, and Video Generator All-in-One)"

x-algorithm. Algorithm powering the For You feed on X

AEON-7. Profile repo — categorized index of NVFP4 model releases, DGX Spark inference stacks, Apple Silicon MLX builds, the voice-AI stack, and the AEON Media Production toolchain · ☕ Tips welcome

ComfyUI. The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface.

crewAI. Framework for orchestrating role-playing, autonomous AI agents. By fostering collaborative intelligence, CrewAI empowers agents to work together seamlessly, tackling complex tasks.

awesome-mcp-servers. A collection of MCP servers.

Gemma-4-26B-A4B-it-Uncensored-NVFP4. NVFP4 Gemma-4 26B-A4B MoE for DGX Spark — optimal recipe: DFlash n=10 (flex) on AEON vLLM Ultimate. 144 tok/s single / 1,724 peak (Coding), up to 158 single (Extraction); beats prior v2 by 20-50%.

neon-serpent. A tiny open-source neon snake game for the browser.

hermes-world-landing. Public landing page for hermes-world.ai — premium dark-fantasy Agentic MMO marketing site.

beautiful-html-templates. A library of HTML slide templates designed so any coding agent can pick the right one and produce a beautiful deck on the user's behalf, automatically.

coolify. An open-source, self-hostable PaaS alternative to Vercel, Heroku & Netlify that lets you easily deploy static sites, databases, full-stack applications and 280+ one-click services on your own servers.

500-AI-Agents-Projects. The 500 AI Agents Projects is a curated collection of AI agent use cases across various industries. It showcases practical applications and provides links to open-source projects for implementation, illustrating how AI agents are transforming sectors such as healthcare, finance, education, retail, and more.

open-design. 🎨 The open-source Claude Design alternative. 🖥️ Local-first desktop app. 🖼️ Your coding agent becomes the design engine: prototypes, landing pages, dashboards, slides, images & video — real files, HTML/PDF/PPTX/MP4 export. 🤖 Claude Code / Codex / Cursor / Gemini / OpenCode / Qwen & 20+ CLIs via BYOK.

docuseal. Open source DocuSign alternative. Create, fill, and sign digital documents ✍️

openshorts. Free & open source AI video platform — Clip Generator, AI Shorts (UGC with AI actors) & YouTube Studio. Self-hosted, no watermarks.

Recordly. Create polished demo videos without editing skills. Mac/Windows/Linux

claude-obsidian. Self-organizing AI second brain for Obsidian + Claude Code. Drop any source and Claude reads, links, and files it into one connected knowledge graph of plain Markdown you own. AI note-taking, personal knowledge management (PKM), and an open-source Notion alternative. Based on Karpathy's LLM Wiki pattern.

openclaw-wiki-recipe. Python

hermes-workspace. Native web workspace for Hermes Agent — chat, terminal, memory, skills, inspector.

claude-mem. Persistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More

cua. Scale computer-use 2.0 with open-source drivers, cross-OS fleets, and benchmarks for training, evaluation, and data generation.

andrej-karpathy-skills. A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls.

gbrain. Garry's Opinionated OpenClaw/Hermes Agent Brain

gstack. Use Garry Tan's exact Claude Code setup: 23 opinionated tools that serve as CEO, Designer, Eng Manager, Release Manager, Doc Engineer, and QA

personal-context-portfolio.

agency-agents. A complete AI agency at your fingertips - From frontend wizards to Reddit community ninjas, from whimsy injectors to reality checkers. Each agent is a specialized expert with personality, processes, and proven deliverables.

ECC. The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

openscreen. Create stunning demos for free. Open-source, no subscriptions, no watermarks, and free for commercial use. An alternative to Screen Studio.

skills. Skills for Real Engineers. Straight from my .agents directory.

DeepTutor. DeepTutor: Lifelong Personalized Tutoring. https://deeptutor.info/.

markitdown. Python tool for converting files and office documents to Markdown.

clicky. Swift

clicky. A proactive tutor mode for Clicky

amp-contrib. Curated skills, tools, MCPs and commands for Amp

ralph. Ralph is an autonomous AI agent loop that runs repeatedly until all PRD items are complete.

antfarm. Build your agent team in OpenClaw with one command.

CodexBar. Show usage stats for OpenAI Codex and Claude Code, without having to login.

nanoGPT. The simplest, fastest repository for training/finetuning medium-sized GPTs.

nanochat. The best ChatGPT that $100 can buy.

obsidian-releases. Community plugins list, theme list, and releases of Obsidian.

paperclip. The open-source app everyone uses to manage agents at work

firecrawl. The API to search, scrape, and interact with the web at scale. 🔥

opencode. The open source coding agent.

last30days-skill. AI agent skill that researches any topic across Reddit, X, YouTube, HN, Polymarket, and the web - then synthesizes a grounded summary

hermes-agent. The agent that grows with you

oh-my-claudecode. Teams-first Multi-agent orchestration for Claude Code

oh-my-codex. OmX - Oh My codeX: Your codex is not alone. Add hooks, agent teams, HUDs, and so much more.

VibeVoice. Open-Source Frontier Voice AI

autoresearch. AI agents running research on single-GPU nanochat training automatically

claw-code. An agent-managed museum exhibit, built in Rust with Gajae-Code / LazyCodex — developed and maintained with no human intervention.

clawchief. Turn your OpenClaw into a Chief of Staff