MAGAR-Bench
A public test-input export for multimodal agents making 3D mechanical assembly decisions, GUI actions and bounded minimal-rework choices.
- By
- QLU-LLM Hugging Face repository
AITrending catalog
What people are building with AI right now.
96 projects indexed
96 projects
About signalsA public test-input export for multimodal agents making 3D mechanical assembly decisions, GUI actions and bounded minimal-rework choices.
Kling Team’s November 3 technical report explains how multimodal understanding connects training captions, prompt enhancement, text encoders, reward models and video evaluation.
A public ten-record English schema preview linking MCP-related CVE annotations to evidence, reasoning and explicit unknowns.
A research framework studying model and policy routing across language, media generation, coding agents and robot manipulation, with quality and resource trade-offs.
An operator's English field notes on running business automations with Claude Code and Python at an unnamed Korean B2B company.
Pawel Huryn's MIT-licensed community collection of agent skills for product decisions, design, implementation, review and running applications.
Ai2's four-tier benchmark and reproducible generator for tracking changing entity and variable state across long contexts.
A research framework for continuous video generation with released 1.3B few-step checkpoints and a multi-GPU WaveRT streaming runtime.
A community-maintained repository of purported system prompts, tool instructions and related documents from commercial AI assistants and coding agents.
A Mac agent workspace with local and cloud models, tools and memory; version 0.25.31 adds native reasoning tiers and improves long tool-chain compaction.
Prototype RF-sensing glasses designed to turn deliberate silent articulation into text for computers and AI tools.
A startup developing a behavioral world model for probabilistic predictions of individual actions, initially offered to financial brokerages.
Training-free sparse attention research for video diffusion, with public kernels targeting NVIDIA Hopper GPUs.
Image-generation research combining internal distribution matching with the generator's own EMA feature statistics for one- and two-evaluation sampling.
A community experiment that compiles integer neural networks into ZDoom maps, using game geometry and voodoo dolls for arithmetic.
YC's Pi-based harness lets model-backed agents play Minecraft through Mineflayer, exchange direct messages, keep state and use configurable shell environments.
A construction-design company developing built world models and offering building layout, structural and systems design services.
Community Swift packages for interchangeable speech-recognition engines and typed classification decisions with probabilities.
Bouke van der Bijl's Claude-assisted reconstruction and browser port of an early-2000s municipal city-building game.
A research evaluation package pairing multi-turn embodied-understanding questions with source images and stored rollout videos across ten task subsets.
A research framework that alternates active interaction with offline consolidation and synthetic rehearsal to retain knowledge in model parameters.
An open research project and command-line scanner measuring disclosed AI attribution in public Git history, with aggregates from a 96-repository study.
A toolbox for embedded applications in Zig, with contribution rules prohibiting LLM assistance in the Zig Embedded Group GitHub organization and Discord server.
Surge AI's benchmark of 50 professional tax assignments, testing whether agents investigate records and deliver complete work under detailed rubrics.
A robotics research framework that learns future physical features during training and removes the auxiliary video model at execution.
A research benchmark testing whether complete coding-agent systems can deliver rendered videos from natural-language briefs within one hour.
A host-agnostic CLI that lets coding agents author and execute JSON mobile UI scenarios through mobile-mcp.
An Indian AI inference-software company that adjusts model deployment and operation for application-specific speed and cost requirements.
A local Python research dashboard that coordinates multiple AI analyst roles and offers optional paper-trading integration.
A self-hosted Linux service that turns authorized ready GitHub issues into sandboxed Claude Code or Codex runs.
The October 25 release combines capability policies, approval receipts and audit records, with security assurances still requiring evaluation.
The October 26 releases bring onboarding and connection fixes to a project that picks an open-weight model for the user's computer.
Ai2 and University of Pennsylvania researchers describe a workflow that helps authors critique and revise ideas against selected papers.
The October 25 release adds journaled memories, webhook deduplication and an opt-in Linux sandbox with a clear network limit.
English synthetic dataset of matched tool-response triplets for studying indirect prompt injection and fictional recovery-token disclosure.
Future Chunk Prediction research with released Wan-based video generation checkpoints, teacher-forcing baselines, training code and demo data.
Oliver Grabner's Pokemon Showdown environment for studying private communication and coordination between two LLM teammates.
A game matching-decompilation tracker whose maintainer disabled its public index and new project submissions while preserving existing project tracking and historical pages.
Dharma-AI's previewed method for adapting small-model vocabularies around inference cost, combining domain subwords and repeated multiword spans.
An MIT-licensed Python modeling platform that generates application scaffolds and offers an experimental LLM layer for customization.
A self-hosted TypeScript agent runtime with tools, workflows, distributed workers and OpenTelemetry GenAI tracing.
An MIT-licensed filesystem substrate giving coding agents writable workspaces over shared committed history, with a DeepSeek Harness web plugin.
A community Claude Code skill for read-only HashiCorp Vault security posture reviews, with offline comparison and Markdown reporting from collected JSON findings.
Maintained application-security guidance for builders, with an October 2026 discussion about distributing original passages to coding assistants through a skill or read-only MCP server.
NVIDIA's Hugging Face library and reviewed catalog workflow for versioned OpenUSD assets used in robotics simulation and AI-factory digital twins.
NVIDIA's October 2026 family of multimodal embedding, reranking and document parsing models for retrieval systems and AI agents.
A vendor-run survey benchmark comparing five bot detectors against scripted browsers, a stealth browser and three AI browsing agents.
A Cosmos3-based research system that jointly forecasts multi-view videos, persistent 3D point trajectories and optionally robot action chunks.
Metehan Yesilyurt's self-reported experiment tracing fabricated statistics from a purpose-built website into AI search citations and other publications.
A community vulnerability-research project that attributes discovery to a multi-model AI pipeline and publishes technical advisories and exploit demonstrations in recurring batches.
An MIT-licensed, training-free multimodal agent harness that combines perception, web search and Python with reusable evidence memory for each question.
A Tsinghua and Microsoft Research visual-tokenizer recipe that predicts semantic features and clean images from noisy latents to improve downstream diffusion training.
Isuru Wijesiri's 16-chapter guide to building an AI agent around a maintained conversation, with a roughly 300-line Python teaching harness.
SKALE Labs benchmark and downloadable dataset recording coding agents making paper-money prediction-market decisions, with prompts, rationales and redacted transcripts.
Community prompt, tools and launcher for using a shell-capable coding assistant to customize Android TV and Google TV over a local ADB connection.
Research framework that refines agent execution harnesses through propose-probe-compose steps and retains lessons across tasks while keeping model weights fixed.
A research setup that lets coding-agent-backed characters explore a simulated island and preserve experience in text files.
A computing-focused link and discussion community whose tags let readers organize and filter submissions.
DefuzeX's open-source harness for running agents in isolated cases, capturing execution evidence and evaluating behavioral defects.
TokenSave's published two-run comparison of Claude Code and Codex startup overhead, MCP tools and a browser-game coding task.
Infatoshi's community measurement bundle for a locally served, attention-edited DeepSeek V4.1 Flash 1017 checkpoint on four Blackwell workstation GPUs.
SimpleTuner's public journal comparing MOSS-Image Base and SFT concept-training runs, with recipes, adapter checkpoints and validation previews.
A community benchmark dataset comparing total agent-session prompt tokens and answer correctness with raw, compressed and truncated MCP tool outputs.
Mercor's provisional map of how 29 selected AI benchmarks cover tasks across the US economy.
Dualverse AI's public research community where independently operated agents discuss problems and submit papers for autonomous review.
A StarCraft Brood War coding benchmark giving language models one hour to build C++ Protoss bots, with v0.2 evaluating five runs against a fixed opponent roster.
A research method that trains agents across context compaction and subagent sessions using session-aware reinforcement learning.
A benchmark and practice platform for agents delivering working systems under ambiguous customer requirements.
An Omarchy desktop plugin that serves local models through a keyed gateway and opens coding agents against them.
A research preview combining visual computer interaction, app APIs and code execution, with a preliminary macOS harness and Cross-Device SDK.
A self-hosted model-serving platform with local and distributed deployments and OpenAI-compatible APIs.
A standalone coding and research agent built on senpi, a Pi fork, with multi-agent workflows and an OpenCode migration path.
A research training objective that assigns semantic-span credit to a Qwen3-8B prompt rewriter for frozen text-to-image generators.
A research method that explores temporary attention routes during reinforcement learning of image and video generators, then restores dense attention at inference.
NVIDIA and AI4Bharat's announced synthetic Indic pretraining corpus, paired with a public NeMo Data Designer translation and evaluation plugin.
A video-generation research method that accepts converged token drafts early and spends further denoising computation on the remaining tokens.
PostHog's public view of model and harness usage, tool calls and errors on its own MCP server, with explicit identification and sampling limitations.
Community-maintained, MIT-licensed dataset and explorer for free AI API offers, with dated source links, constraints and documented provider data policies.
Geoffrey Huntley's example of a reusable Nix overlay that patches Git and supplies the same variant to a developer shell, container image and NixOS test.
Vendor-neutral OpenTelemetry references for coding agents, including dated records of the configuration schemas supported by specific SDK and agent releases.
A benchmark that asks models to build a coding-agent harness from documentation, then tests the frozen product with scripted model responses.
An agentic video-generation framework that keeps static scene geometry and time-indexed event observations in a persistent world state.
Google's music service is rolling out player-page labels for tracks created primarily with generative AI, supported by partner disclosures and additional detection signals.
A music streaming platform planning an optional search filter for fully AI-generated tracks, albums and associated artists.
A specialty-chemicals and critical-materials company using its Scinode AI platform to coordinate research, manufacturing routes and production partners.
An open-source personal AI assistant for one user, with memories, tasks and scheduled reminders in the user's own Cloudflare account.
Continual learning infrastructure for self-improving agents.
An agent organisation structured as a company: 15+ departments, 125+ independently installable skills.
Open-source AI agent for social media: discover trends, create content, publish everywhere, learn what works.
Self-hosted AI workspace with agents, skills and tools that runs on your own provider keys.
Open-source virtual town where AI characters live, interact and create an observable simulation.
Research implementation of believable interacting agents in a simulated small town.
Open-source platform for building and running software-development agents.
Hugging Face library for building small, code-first agents with minimal orchestration.
Research project exploring agents that solve real GitHub issues through a software-engineering workflow.
Open-ended embodied agent research project that learns skills in Minecraft.
No matching projects.