Newly Accepted

A DeepMind technical report PDF about double-blind AI evaluations and confidentiality issues in AI safety research.

A research blog post describing an open-source small transformer trained from scratch to reach 44% on ARC-AGI-1 at very low cost, with architecture notes, ablations, and pointers to code.

Anthropic’s release notes for Claude Fable 5.1, covering new features, breaking changes, pricing, model availability, and migration details for API users.

Open-source platform for building internal tools, dashboards, business apps, and workflows, with AI app generation and MCP support for coding agents.

Supercov is a coverage tool for coding agents that turns uncovered paths into the next test to write, then reruns the suite to verify progress.

Subpool is a self-hosted control plane for managing AI subscription and API capacity across teams. It pools authorized accounts, issues employee-specific keys, exposes OpenAI-compatible endpoints, and tracks usage without storing prompts or responses.

Cupertino is a macOS app that exposes Apple apps like Mail, Calendar, Contacts, Messages, Notes, Safari, Maps, Screen, and Sound through MCP servers for agents. It includes a signed menu-bar app, a visible permission model, read/write controls, and a live session log.

HEIDES is a deterministic local harness for AI coding agents that maps a codebase, grounds plans against the current graph, and checks proposed patches for conflicts, edge cases, and security issues before changes land.

A local zero-dependency sandbox for agents that records files, commands, database activity, network access, and screenshots for review before approving changes.

Shed is a GitHub-hosted CLI for managing read-only repo catalogs and disposable writable workspaces for terminal coding agents. It syncs fresh checkouts, supports parallel agent sessions, and integrates with Claude Code, Cursor CLI, and opencode.

A local-first authority layer for Markdown used by humans and AI agents to manage competing project documents, current sources, and review workflows.

An Mr. T–themed skill that tells an AI to skip preamble and answer directly. The repo shows the purpose, example phrases, and a simple installation step via SKILL.md.

Google’s Gemini 3.8 Flash model page for developers, describing a high-speed multimodal model with support for agents, function calling, structured outputs, code execution, file search, computer use, and long context.

A production-ready Rails SaaS starter designed to give coding agents strong architectural context, with authentication, billing, authorization, Hotwire, Tailwind, Kamal, and repo-based guidance for extending the app consistently.

A serverless reverse geocoding SDK and browser experience for web and embedded devices, with support shown for WebMCP and embedded targets like RP2040, RP2350, and ESP32.

Live demo of a prompt-injection defense for LLMs using per-position semantic overlays. The page shows how marked spans can be treated as untrusted while the base model stays frozen, with interactive examples for retrieved text, hidden instructions, language marking, and request-scoped overlays.

AI-powered project collaboration workspace for teams to manage work items, decisions, blockers, and context in one shared place. It includes AI-suggested updates, provider support, MCP connections for agents, an API/CLI, and self-hosting via Docker Compose and PostgreSQL.

A CLI tool that exposes a local MCP server through a public URL with one command. It supports anonymous ephemeral tunnels, OAuth 2.1 by default, and a self-hosted relay option.

A source-linked public archive of documented incidents involving AI coding agents and chatbots, with summaries, lessons learned, and an incident index.

An MCP-based research agent that searches live web data through SQL and publishes reports with the full query and tool trajectory.

An interactive quiz from Raq.com that asks users to identify which AI model created each of ten default website designs, with links to the live pages after each reveal.

PearPie is a private AI chat app for desktop and mobile that avoids accounts, stores chats on your own devices, and syncs them peer-to-peer. It supports local models plus premium hosted models, with per-conversation model selection and pay-as-you-go credits.

Open-source Raspberry Pi bird deterrent that uses an AI model to detect birds near a pool and trigger a sprinkler automatically.

Open-source harness for measuring how consistently hosted LLM APIs return the same output across repeated runs and over time, with raw transcripts, scoring, and a static leaderboard generator.