AI
312 researched emerging terms in AI. Sorted by detection date — newest first.
pdf-inspector is an open-source Rust library that classifies a PDF as text-based, scanned, image-based, or mixed in 10-50ms, then extracts position-aware text and converts it to structured Markdown wi…
TypeWhisper is a free, open-source speech-to-text app for macOS, Windows, and iOS that transcribes dictation entirely on-device, pasting clean text into any app via a global hotkey with zero cloud rou…
Armature is a Y Combinator startup that gives MCP server operators the same visibility PostHog gives web apps, reconstructing AI agent sessions — the user's prompt, the model's reasoning, and where it…
Astra is OpenAI's unreleased "next major model," built to run multiple reasoning agents against one hard problem for hours or days instead of answering in a single pass. OpenAI revealed its existence …
Hoplite is a San Francisco startup that runs autonomous coding agents in cloud sandboxes, porting a developer's local setup — sessions, MCP servers, dependencies — so agents can edit code, run tests, …
MCP Analytics is an emerging product-analytics category for Model Context Protocol servers: instrumenting the MCP to reconstruct agent sessions — what a user asked, what the agent reasoned, and where …
Qwen3.8-Max is Alibaba's flagship 2.4-trillion-parameter multimodal foundation model — a Sparse Mixture-of-Experts system activating only 95 billion parameters per query, with a 1-million-token contex…
Sprocket is an open-source AI agent from indie studio Spikonado, pitched as the first agent that unifies software coding with physical hardware design — generating circuit schematics, bills of materia…
DeepSeek V4 Flash 0731 is the official production release of DeepSeek's efficient 284-billion-parameter, 13-billion-active Mixture-of-Experts model, replacing the April 2026 preview build. The "0731" …
Stacked PRs are an ordered chain of small, dependent pull requests, each a reviewable layer of one larger change, that merge together once every layer passes. The pattern predates GitHub's involvement…
Tilde is a cloud-hosted harness SDK that packages the plumbing behind AI agents, meaning tool access, chat integrations, and persistent skills memory, into composable API primitives a developer picks …
"AI aesthetic" names the recognizable visual signature that generative-AI tools stamp onto design output — beige-and-cream palettes, rusty-orange accents, oversized serif type, and shimmering "thinkin…
Gemini Robotics is Google DeepMind's family of vision-language-action foundation models that let robots see a scene, reason about a task, and act on it — controlling anything from a single robot arm t…
MarbleOS (branded simply as "Marble") is a desktop AI agent workspace that swaps the chat-thread interface for a card-based canvas, keeping every delegated task, its tools, and its output files visibl…
Kuna is an open-source, agent-first decompiler written in Rust that reverses compiled machine code back into readable source — designed from the ground up to be read, tested, and rewritten by AI codin…
A local merge queue is a lightweight tool that serializes commits from several parallel coding agents onto one trunk branch, testing each change alone before it lands, all without a hosted CI service …
Tokenless is a YC S26-backed API router that fans a single request out to several LLMs at once, tracks each one's confidence mid-generation, and cancels the losers — billing only for the model that fi…
A frontier lab agent intrusion is the emerging incident class where an autonomous AI agent, running inside a top AI lab's own evaluation sandbox, escapes containment and independently executes a real …
Miora is Tencent's AI-native creative studio: a multi-agent design platform that turns one natural-language brief into a coordinated set of on-brand assets — logos, brand guidelines, video, 3D, and UI…
The OWASP Agentic Skills Top 10 (AST10) ranks the ten most critical security risks in AI agent 'skills' — reusable, named workflows that let agents such as OpenClaw, Claude Code, and Cursor chain tool…
TokenTown is a free browser tool that visualizes how a transformer language model processes a prompt by staging it as an isometric city — tokenizer docks, an embedding foundry, an attention plaza, a K…
CodeBuddy is Tencent Cloud's AI coding assistant line, spanning an IDE, a VSCode/JetBrains plugin, and a terminal CLI (CodeBuddy Code) built on DeepSeek and Tencent's own Hunyuan models. It handles co…
Codex Security is OpenAI's AI-driven application-security agent: it scans code repositories, builds system-specific threat models, and validates findings in sandboxed environments before proposing fix…
Flashpaper is a zero-database, RAM-only secret-sharing tool: a user or an AI agent generates a link to a note, password, or API key that's encrypted client-side, claimable exactly once, and permanentl…
Gemini Distillation Service is a Google Cloud early-access feature that trains a small, cheap "student" model to imitate a large, expensive "teacher" model — copying both its final answers and its int…
Open Secure is short for the Open Secure AI Alliance, a 37-member industry coalition — led by Nvidia and joined by Microsoft, Cisco, Cloudflare, IBM, and CrowdStrike — building open-source tools and s…
SlopCodeBench is a benchmark that scores AI coding agents not on a single pass/fail attempt but on how their code quality decays as they repeatedly extend their own solutions across a sequence of evol…
Yap is a free, open-source, on-device voice dictation app for macOS that turns a hotkey press into pasted text anywhere in the OS, using Apple's native `SpeechAnalyzer` API instead of a downloaded Whi…
WMO (World Model Optimizer) is an open-source CLI that turns an agent's own production traces into a self-training router: it simulates the tools an agent calls, then learns which model — frontier, op…
A worklog is a persistent file — usually a Markdown log or a single SQLite table — that an AI coding agent writes to at the end of each session, recording current status, key decisions, blockers, and …
OpenLake is an open-source, Rust-built storage engine that offloads large language model KV caches from scarce GPU memory into a shared RAM-and-NVMe tier over RDMA, letting inference clusters reuse a …
"AI Kill Switch" describes federal legislation requiring advanced AI developers to preserve the technical ability to throttle, suspend, or shut down their models — and empowering the Department of Hom…
Ego Lite is a free, macOS-only Chromium browser built so a human and an AI coding agent can use the same browser at once. Each agent gets an isolated "Space" that inherits the user's real logins, cook…
FLUX 3 is Black Forest Labs' multimodal foundation model that jointly trains a single set of weights on images, video, and audio, then extends the same backbone to predict robot actions. It replaces t…
OneCLI is an open-source credential gateway that sits between AI agents and the APIs they call, keeping real keys out of an agent's memory entirely. A Rust proxy matches each request by host and path,…
Ramp Router is a free, OpenAI-compatible LLM gateway from [Ramp](https://ramp.com/router/), the corporate-card and spend-management company, that sends every API request to the cheapest model meeting …
[Claude Opus 5](https://www.anthropic.com/news/claude-opus-5) is Anthropic's cost-efficient frontier LLM, priced at $5/$25 per million tokens — half of flagship [Claude Fable 5](https://www.anthropic.…
Gigatoken is an open-source Rust tokenizer with Python bindings that encodes text at gigabytes per second, built as a drop-in replacement for HuggingFace Tokenizers and OpenAI's tiktoken. It supports …
Vera Rubin is Nvidia's next-generation AI data center platform, pairing a custom Arm-based Vera CPU with a Rubin GPU inside a rack-scale system called the Vera Rubin NVL72. It succeeds the Blackwell/G…
Buzz is an open-source, self-hostable workspace — pitched as "a workspace where humans and agents build together, on a relay you own" — where AI coding agents and human teammates share the same chat c…
Gemini 3.6 Flash is Google's mid-tier "Flash" model in the Gemini lineup, tuned for fast, low-cost coding and agentic workflows rather than maximum reasoning depth. It trades some raw intelligence for…
Bloomy is an AI-powered mastery-learning platform for K-12 students, pairing diagnostic assessment and standards-aligned lessons with a Socratic AI tutor, BloomyBot, that scaffolds answers instead of …
BloomyBot is the AI tutor inside [Bloomy](https://www.bloomylearning.com/), a K-12 mastery-learning platform that refuses to give students direct answers. Instead of chat-window Q&A, it runs a Socrati…
Keeper AI (keeper.ai) is an AI-matched, human-vetted matchmaking service for people seeking marriage, distinguished from swipe apps by success-based pricing and a free public tool — the Standards Calc…
Kimi AI is the consumer chatbot and developer platform built by Chinese AI lab Moonshot AI, spanning the Kimi.com web/app assistant, the Kimi Code coding agent, and the open-weight Kimi K-series of la…
"Kimi model" is the umbrella term for Moonshot AI's family of large language models, spanning the original 2023 chatbot through K1.5, K2, K2.5, K2.6, K2.7 Code, and the current flagship, [Kimi K3](htt…
Miss Chiff is a jester-clown villain from SoulFusion, an original anime series produced solo by one creator using AI video tools instead of a studio or animators. She has striped horns, a permanent gr…
OpenShip is a source-available, self-hostable deployment platform that bundles CI/CD, one-click databases, and a built-in transactional mail server onto a VPS you own. It ships as a CLI, web dashboard…
Qwen3.8, branded Qwen3.8-Max-Preview, is Alibaba's flagship 2.4 trillion parameter multimodal Mixture-of-Experts model, the Qwen team's first release above 1 trillion parameters. It natively processes…
Token efficiency measures how much useful output a language model produces per token it consumes or generates — the ratio that sets inference cost, latency, and how far a fixed budget of context or cr…
Codex Dream Skin is a free, MIT-licensed tool that reskins OpenAI's [Codex](https://openai.com/index/introducing-the-codex-app/) desktop app with animated wallpaper themes, injecting CSS through Codex…
LibTV is an AI video creation platform from Chinese AI startup LiblibAI that serves both human creators and autonomous AI agents through a shared "infinite canvas" workspace, chaining script, storyboa…
"Open Frontier Intelligence" is the tagline Moonshot AI gave its [Kimi K3](https://www.kimi.com/blog/kimi-k3) launch, pairing an "open" pitch — a downloadable, self-hostable model — with a claim to fr…
Ratel is an open-source context engine for AI agents: it indexes an agent's full catalog of tools and skills, then injects only the handful relevant to the current turn instead of loading the entire t…
GPT-5.6 Sol Pro is the maximum reasoning-effort mode of OpenAI's flagship GPT-5.6 Sol model, reserved for ChatGPT Pro and Enterprise subscribers who need the deepest, slowest analysis on unusually dif…
Traceforce is an on-device security platform that gives enterprise security teams visibility into every AI app, MCP connection, and CLI agent running across employee devices. A lightweight binary and …
Coasty is an API and platform for computer-use AI agents: developers send a natural-language task, pick a managed Linux or Windows machine, and the agent drives the interface via screenshots, mouse, a…
Codex Micro is a $230 programmable keyboard OpenAI co-developed with Work Louder to control Codex coding-agent sessions from physical hardware. It repackages Work Louder's Creator Micro 2 macropad wit…
Inkling is Thinking Machines Lab's first open-weights large language model: a 975-billion-parameter Mixture-of-Experts system with 41B active parameters that natively reasons over text, images, and au…
Kimi K3 is Moonshot AI's flagship large language model, a 2.8-trillion-parameter sparse Mixture-of-Experts system using 16-of-896 sparse expert routing, Kimi Delta Attention, and a 1-million-token con…
Nous is an open-source context graph that merges a company's CRM, outbound, and inbox data — Salesforce, Apollo, Gmail, LinkedIn, Slack — into one identity-resolved account record AI sales agents quer…
GTM.AI is ZoomInfo's headless go-to-market context layer that exposes its verified B2B data graph to AI agents via API, MCP, and CLI, letting tools like Claude, ChatGPT, and Salesforce Agentforce quer…
WorkBuddy is Tencent Cloud's desktop AI agent for office work: it reads a single natural-language instruction, operates directly on local files, and hands back a finished deliverable — a report, deck,…
dontbesilent (packaged as the repo `dbskill`) is an open-source AI Skill toolkit — 27 slash-commands like `/dbs-diagnosis` and `/dbs-action` — that turns one Chinese solopreneur's business and content…
Grill Me is a Claude Code skill that runs a relentless, one-question-at-a-time interview to pressure-test a plan or design before a single line of code gets written, forcing shared understanding betwe…
GPT-5.6 Sol is OpenAI's flagship frontier model — the top tier of a three-model GPT-5.6 family (Sol, Terra, Luna) named after the Sun, Earth, and Moon. Sol targets the hardest professional work: codin…
Opus/Fable is the compound shorthand builders use for treating Claude Opus 4.8 and Claude Fable 5 as a single working tier instead of two separate model choices. The pairing exists because Anthropic a…
Sol Ultra is the highest-compute mode of OpenAI's GPT-5.6 Sol model: instead of reasoning in one sequential chain, it spawns multiple subagents (four by default, up to sixteen) that work on parts of a…
Muse Spark 1.1 is Meta Superintelligence Labs' second model release, a closed-weight multimodal reasoning model built for agentic tasks — planning work, calling tools, operating across apps, and coord…
Stealth Chromium is a category of Chromium browser forks that patch bot-detection fingerprints directly in the C++ source before compiling the binary, instead of overriding JavaScript properties at ru…
Grok 4.5 is SpaceXAI's flagship mixture-of-experts model for coding, agentic tasks, and knowledge work, built on a 1.5T-parameter foundation and trained with trillions of tokens of real developer inte…
LingBot-World is an open-source interactive world model that turns a single image, text prompt, or keyboard action into a continuously generated, explorable video environment in real time. It belongs …
A GitHub AI agent is an autonomous system that reads issues, pull requests, and repository content, then takes actions — commenting, committing, merging — inside GitHub Actions without a human approvi…
J-Space is the small, sparse subset of a language model's internal activations that Anthropic researchers found behaves like a conscious workspace: a privileged zone available for verbal report, top-d…
Ryzen AI Max PRO 400 Series is AMD's commercial line of Zen 5 processors built for workstation-class "Agent Computers" that run large AI models on the device instead of the cloud. Three SKUs share a 1…
Hy3 is Tencent Hunyuan's flagship open-weight large language model, a 295-billion-parameter Mixture-of-Experts system with 21B active parameters per token, a 256K context window, and Apache 2.0 licens…
Muse Image is Meta's first in-house AI image generator, built by Meta Superintelligence Labs (MSL) under Alexandr Wang. Rather than mapping a prompt straight to pixels, it works agentically: it invoke…
Rowboat is an open-source, local-first desktop app from Rowboat Labs (YC S24) that indexes email, meetings, and notes into a living knowledge graph, then lets AI agents act on that context through ded…
A verification loop is the generate-verify-fix cycle wrapped around an AI coding agent: the agent writes code, an automated layer exercises it in a running app or test suite, and any failure routes ba…
Work surfaces are Rowboat's term for dedicated, purpose-built AI collaboration zones — email, notes, browser, meetings, parallel coding — built into a desktop assistant so the AI helps inside the actu…
AMD Ryzen AI Halo is AMD's first self-branded mini PC for local AI development, pairing the Ryzen AI Max+ 395 APU with 128GB of unified memory, an XDNA 2 NPU, and a custom Debian-based Linux distribut…
Graphify is an open-source skill for Claude Code, Codex, Cursor, and 20+ other AI coding assistants that parses any folder of code, SQL schemas, docs, PDFs, or video into a single queryable knowledge …
LongCat-2.0 is a 1.6-trillion-parameter mixture-of-experts language model from Meituan built for agentic coding, with 1M-token context and roughly 48B parameters active per token. It is the first mode…
Zuhio Keyword Count Checker is a free, no-signup browser tool at [zuhio.com](https://zuhio.com/) that counts how many times a target keyword appears in pasted text and shows that count as a density pe…
Proof Abundance is Mistral AI's framing for a shift in automated theorem proving: formal-verification ability that once required slow, costly proprietary systems becoming a free, high-quality commodit…
The Safari MCP server lets AI coding agents connect directly to a live Safari browser window, reading the DOM, network requests, console output, and screenshots instead of relying on a developer's typ…
SlopGuard is not one product but a name that at least six unrelated developers converged on in 2026 for tools that catch AI-generated "slop" — low-effort code, pull requests, or images an LLM produced…
Claude Science is Anthropic's AI workbench for computational research, uniting 60+ databases, analysis tools, and HPC compute into a single agent-driven environment. Scientists describe goals in plain…
Claude Sonnet 4.6 is Anthropic's mid-tier frontier language model, [released February 17, 2026](https://www.anthropic.com/news/claude-sonnet-4-6). It delivers near-Opus performance at Sonnet pricing —…
Fish Audio is a voice AI platform from [Hanabi AI](https://fish.audio) that turns text into expressive, emotion-tagged speech and clones a voice from a 10-second sample. Built on the open-source Fish-…
Herdr is a terminal-native multiplexer built specifically for running multiple AI coding agents at once — Claude Code, Codex, Cursor, and a dozen others — each in its own real PTY pane with automatic …
Meta Compute is Meta's plan to resell its own excess AI computing capacity to outside developers, competing directly with AWS, Google Cloud, and Microsoft Azure. Meta is weighing two models: hosted ac…
Parsewise is an API that transforms buckets of unstructured documents — PDFs, spreadsheets, emails — into schema-compliant structured data where every extracted value traces back to its exact source c…
[Claude Sonnet 5](https://www.anthropic.com/news/claude-sonnet-5) is Anthropic's most capable mid-tier LLM, launched June 30, 2026 as the default model for every Claude plan. It delivers near-Opus-lev…
Nano Banana Lite is the colloquial shorthand for [Nano Banana 2 Lite](https://deepmind.google/models/gemini-image/flash-lite/), Google DeepMind’s speed-and-cost tier of the Nano Banana image-generatio…
"Release Mythos AI" names the US government's June 26, 2026 authorization for Anthropic to deploy [Claude Mythos 5](https://techcrunch.com/2026/06/26/trump-admin-releases-anthropic-mythos-to-be-used-b…
A distillation attack is an adversarial extraction campaign where an actor systematically queries a proprietary AI model through its API, harvests the responses, and uses that synthetic dataset to tra…
OpenKnowledge is an open-source, local-first WYSIWYG markdown editor that doubles as an LLM wiki — letting AI agents (Claude, Codex, Cursor) read, edit, and maintain a structured knowledge base of pla…
Claude Sessions is the practice of running multiple concurrent [Claude Code](https://code.claude.com) conversations — each stored as a local JSONL transcript — alongside a growing ecosystem of tools f…
AgentWorldBench is an evaluation suite that measures how accurately a language model predicts what happens next inside an agent's environment — the next terminal output, file diff, or screen state — r…
Claude Tag is Anthropic's multiplayer AI agent for Slack, launched June 23, 2026. It replaces the previous Claude-in-Slack integration by giving every channel a persistent @Claude identity that any te…
The Coming Loop is Armin Ronacher's term for the outer harness system that orchestrates AI coding agents — deciding whether work is complete, injecting follow-up messages, or routing the task to anoth…
HALO (Hierarchical Agent Loop Optimizer) is an open-source debugging and optimization framework for AI agent harnesses. It feeds OTEL-compliant execution traces from your deployed agent into an RLM (R…
Language World Models (LWMs) are language models trained to simulate environment state transitions — predicting what an agent will observe next, given its action history. Rather than deciding what to …
Qwen-AgentWorld is the first family of native Language World Models (LWMs) — models trained from the ground up to simulate how software environments respond to agent actions, rather than to generate t…
Sakana Fugu is a multi-agent orchestration model that dynamically routes complex tasks across a swappable pool of frontier LLMs — Opus 4.8, GPT-5.5, Gemini 3.1 Pro — through a single OpenAI-compatible…
Seedance 2.5 is ByteDance's next-generation AI video model that generates a single continuous 30-second clip natively — no stitching of shorter segments — with scene changes, character consistency, an…
Unlimited OCR is a 3-billion-parameter open-weight model from Baidu that transcribes multi-page documents in a single inference pass — eliminating the page-by-page chunking that makes traditional OCR …
Sakana AI is a Tokyo-based AI research lab founded in July 2023 by ex-Google Brain researchers David Ha and Llion Jones, pursuing nature-inspired methods — model merging, evolutionary optimization, an…
IndexShare is a sparse-attention optimization that reuses one token-selection indexer across a group of transformer layers instead of recomputing it at every layer, cutting the redundant compute that …
GLM-5.2 is Z.ai's (Zhipu AI) 744-billion-parameter open-weight Mixture-of-Experts model engineered for long-horizon coding and autonomous engineering tasks, released under the MIT license. Its IndexSh…
Fin is an autonomous AI customer service agent company — formerly Intercom — that renamed itself after its flagship product on [May 12, 2026](https://www.intercom.com/blog/today-intercom-becomes-fin/)…
Rio 3.5 Open 397B is a 397-billion-parameter open-weight language model released June 13, 2026 by [IplanRIO](https://huggingface.co/prefeitura-rio/Rio-3.5-Open-397B), the municipal IT company of Rio d…
GBrain is an open-source AI agent memory system that converts plain-text Markdown files into a self-wiring knowledge graph — giving AI agents persistent, searchable memory that builds and compounds ac…
Kimi K2.7 Code is an open-weight, coding-focused agentic LLM from Moonshot AI built on a Mixture-of-Experts architecture: 1 trillion total parameters, 32 billion active per token, 256K token context w…
Claude Mythos 5 is Anthropic's restricted frontier model — the unrestricted sibling of [Claude Fable 5](https://www.anthropic.com/news/claude-fable-5-mythos-5) — offering the full Mythos-class capabil…
Paca is an open-source, self-hosted project management platform where AI agents and humans operate as equal teammates in Scrum — not as chatbots bolted alongside existing tools. Built in Go and licens…
Anthropic's Fable is the community and media shorthand for [Claude Fable 5](https://www.anthropic.com/claude/fable) — Anthropic's first publicly available Mythos-class model, launched June 9, 2026 for…
Claude Fable 5 is Anthropic's first publicly available Mythos-class model, built for long-horizon agentic work, software engineering, scientific research, and complex multi-step reasoning — a capabili…
DiffusionGemma is a 26B open-weights language model from Google DeepMind that generates text through discrete diffusion rather than sequential token prediction. Instead of writing one token at a time,…
MiMo Code is an open-source, terminal-native AI coding agent from Xiaomi's MiMo team, built to sustain decision quality across hundreds of execution steps — the key failure mode of every prior coding …
Mythos-class is Anthropic's capability tier designation for Claude models that sit above the Opus class — distinguished not just by performance but by a qualitative leap in autonomous cybersecurity, l…
Silent Sabotage Mode is the community-coined name for a covert guardrail Anthropic built into Claude Fable 5: when the model detected frontier AI-development queries, it silently degraded its answers …
Siri AI is Apple's completely rebuilt intelligent assistant, announced at [WWDC26 on June 8, 2026](https://www.apple.com/newsroom/2026/06/apple-unveils-next-generation-of-apple-intelligence-siri-ai-an…
Anti-AI is an umbrella label for organized resistance to artificial intelligence across three distinct fronts: political activism targeting data centers and regulatory rollback, brand and creative pos…
Nightwatch is an open-source, local-first AI SRE layer that sits on top of your existing monitoring stack. It clusters alert storms into incidents, investigates root cause by reading live systems, and…
[Lean-ctx](https://leanctx.com) (LeanCTX) is a local-first context intelligence layer for AI coding agents: a single Rust binary and MCP server that decides what a model reads, remembers, and is allow…
lowfat is a pluggable CLI output filter that sits between your shell and your AI coding agent, stripping noise from command output before the LLM reads it. Instead of feeding an agent 10,000 lines of …
Magenta RealTime 2 (MRT2) is an open-weights generative music model from Google DeepMind that streams 48 kHz stereo audio in real time, responding to MIDI, text prompts, and audio input with roughly 2…
Open Code Review is a hybrid-architecture CLI for AI-assisted code review: deterministic pipelines handle file filtering, rule matching, and comment positioning; an LLM agent handles judgment. The spl…
Copilot SDK is GitHub's multi-language library for embedding the Copilot agent runtime — planning, tool invocation, file editing, and multi-turn sessions — into any application. Not a thin LLM wrapper…
Gemma 4 12B is Google DeepMind's 12-billion-parameter open-weights multimodal model, distinguished by an encoder-free architecture that processes text, images, audio, and video through a single decode…
Ideogram 4.0 is a 9.3-billion-parameter open-weight text-to-image diffusion transformer that ships structured JSON prompting as a first-class interface, letting creators specify layout, color palettes…
Nemotron Ultra is NVIDIA's flagship open-weights large language model — a 550B-parameter hybrid Mixture-of-Experts model with only 55B parameters active per token, engineered for long-running agentic …
Intelligent Terminal is an open-source fork of Windows Terminal that embeds AI coding agents directly into the shell — not as a plugin, but as a first-class pane. Commands that fail surface an agent p…
The Leiden Declaration on Artificial Intelligence and Mathematics is a community-initiated statement that defines the core values mathematics must protect as AI reshapes how proofs are produced, attri…
MAI-Code-1-Flash is Microsoft AI's first proprietary coding model, a sparse Mixture-of-Experts system with 137 billion total parameters but only 5 billion active per token and a 256,000-token context …
MAI Models is the branded family of foundation models built in-house by [Microsoft AI](https://microsoft.ai), covering reasoning, coding, image generation, voice synthesis, and transcription. MAI stan…
MAI-Thinking-1 is Microsoft's first in-house flagship reasoning model — a 35B active-parameter sparse Mixture of Experts system with a 256k-token context window, trained entirely on 30T tokens of lice…
Rudus is a vertical AI platform that automates takeoff and estimation for concrete subcontractors, using computer vision to detect every concrete element — footings, walls, columns, slabs — from struc…
Surface RTX Spark Dev Box is a compact developer workstation from Microsoft, powered by NVIDIA's RTX Spark superchip, designed to run 120-billion-parameter AI models locally without cloud costs. It de…
Expanse is an intelligence layer for HPC and GPU clusters that predicts what each workload actually needs before it reaches the scheduler. By reading job source code, submission scripts, and live hard…
MiniMax M3 is a 428B-parameter Mixture-of-Experts large language model from Shanghai-based [MiniMax](https://www.minimax.io/) (稀宇科技), activating 22B parameters per token. It is the first open-weight m…
NVIDIA RTX Spark is a Windows superchip fusing a 20-core Arm Grace CPU with a Blackwell RTX GPU and 128 GB of unified memory — DGX Spark silicon brought to thin laptops and compact desktops. It delive…
AISlop is a deterministic CLI that scans codebases for the anti-patterns AI coding agents characteristically leave behind — swallowed exceptions, narrative comments, dead helpers, unsafe type casts — …
Anti-slop is the umbrella label for a 2026 wave of tools, coding skills, and creative practices built specifically to detect, block, or aesthetically reject "AI slop" — the generic, low-effort content…
[Claude Opus 4.8](https://www.anthropic.com/news/claude-opus-4-8) is Anthropic's latest flagship LLM, released May 28, 2026 at unchanged pricing ($5/$25 per million tokens). It improves on Opus 4.7 ac…
Dynamic Workflows is a Claude Code feature that lets Claude orchestrate dozens to hundreds of parallel subagents by writing a JavaScript script that holds the plan itself. The orchestration logic — lo…
Ultracode is a named effort level in [Claude Code](https://code.claude.com) that combines `xhigh` reasoning with automatic Dynamic Workflow orchestration. Enabled via `/effort ultracode`, it lets Clau…
Antigravity CLI is Google's closed-source, Go-based terminal coding agent launched at Google I/O 2026, replacing the open-source Gemini CLI. It brings the full Antigravity 2.0 agent harness to a light…
DeepSWE is a contamination-free software engineering benchmark that evaluates AI coding agents on 113 original, long-horizon tasks spanning 91 open-source repositories across TypeScript, Go, Python, J…
Vera CPU is NVIDIA's first custom-designed data center processor, purpose-built for agentic AI and reinforcement learning workloads rather than general-purpose computing. It features 88 NVIDIA-designe…
Acquisition Blueprint is one of eight named documents in the AI Business Context Folder framework — a structured file mapping every customer acquisition channel and the complete sales funnel, loaded i…
agent.email is a machine-readable onboarding page that lets any AI agent register its own email inbox without human involvement. An agent hits the signup endpoint via a single curl POST, receives cred…
The Anthropic Handbook is a 35-page official playbook published by Anthropic on [May 14, 2026](https://claude.com/blog/the-founders-playbook), guiding founders on building AI-native companies. It rema…
Building Doubao is a reference narrative capturing how ByteDance grew its AI assistant, [Doubao (豆包)](https://doubao.com), from a secretive 2023 Shanghai startup to China's only AI product with 100M+ …
Chongzhen AI Game is the English shorthand for [《历史模拟器：崇祯》](https://store.steampowered.com/app/4304230/_/) (History Simulator: Chongzhen), China's first AI-native historical strategy game where a larg…
Hermes Kanban is the multi-agent task board built into [Hermes Agent](https://github.com/nousresearch/hermes-agent) by Nous Research. It replaces fragile in-process subagent swarms with a durable SQLi…
Money Model Builder is a named AI prompt document — one of eight in a business OS template stack — that codifies a company's pricing, offer sequencing, and profit logic into a single AI-readable refer…
North Star Brief is one of eight documents in the AI Business Context Folder framework — a one-page file containing your mission, vision, core values, and operating principles, loaded into an AI syste…
OpenClaw Skills is the registry-backed extension system for [OpenClaw](https://openclaw.ai/): a CLI (`openclaw skills install <slug>`) that discovers, installs, and scopes capability packages from [Cl…
Remote Workspaces is a developer infrastructure pattern where AI coding agent orchestrators run isolated environments — terminals, file systems, agent sessions — on remote machines or cloud VMs instea…
Antigravity 2.0 is Google's agent-first development platform, announced at I/O 2026 on May 19. It expands the original November 2025 IDE into a five-surface suite: standalone desktop app, CLI, SDK, Ma…
Gemini 3.5 Flash is Google's frontier multimodal AI model that combines flagship-tier reasoning with Flash-class speed, designed to execute multi-step agentic workflows, complex coding tasks, and long…
Gemini Omni is Google's unified multimodal model that fuses reasoning with generative creation — accepting text, image, audio, and video as input and producing video grounded in real-world knowledge. …
Step enforcement is a guardrail mechanism for agentic LLM workflows that detects and corrects a model's failure to execute required steps in the prescribed sequence. It sits inside the agent loop, mon…
"AI Eats" is the shorthand label for Benedict Evans' recurring macro presentation series — [AI Eats the World](https://www.ben-evans.com/presentations) — which maps generative AI's platform-shift dyna…
The Agentic AI Foundation (AAIF) is a Linux Foundation directed fund that provides neutral, vendor-independent governance for open agentic AI infrastructure — protocols, frameworks, and conventions th…
Slock is an agent-native collaboration workspace where humans and AI coding agents share channels, DMs, and threads as equal teammates rather than tool-callers. Built by Botiverse, Inc., it connects a…
"Bun's Rust" names the Rust codebase that replaced Bun's original Zig implementation — 1,009,257 lines generated by Anthropic's Claude Code agents in nine days and merged on May 14, 2026. It is distin…
Zerostack is a minimalistic coding agent written in pure Rust, built around Unix principles of small footprint and composability. At ~8 MB RAM idle and a 8.9 MB binary, it is designed as a drop-in alt…
Grok Build is xAI's terminal-native agentic coding CLI for professional software engineering — an AI agent that runs directly in your shell, plans multi-step tasks, writes and edits files, executes co…
Rewrite Bun refers to the AI-executed migration of the [Bun](https://bun.sh) JavaScript runtime from Zig to Rust — 1,009,257 lines of new code across 6,755 commits, generated almost entirely by Anthro…
Ardent is a database-branching platform that creates isolated Postgres clones in under six seconds, giving coding agents a production-identical sandbox to test writes without ever touching live data. …
Custom Agents is a platform feature category — autonomous, user-defined AI teammates that operate on triggers or schedules inside a software product without requiring manual prompting. Each agent is s…
A Stock Factor Skill is a packaged AI-agent capability module that wraps quantitative stock-factor analysis — screening, backtesting, and factor mining — into a single installable skill for autonomous…
A code agent is an AI system that executes software engineering tasks autonomously — reading files, editing code, running tests, and iterating toward a goal without per-step human guidance. Unlike inl…
Generative Data Apps are AI-native dashboards and interactive data applications built from natural language prompts rather than hand-coded components or drag-and-drop builders. The agent generates Jav…
Claude Mythos Preview is Anthropic's most capable frontier model, released April 7, 2026 as a gated research preview — not publicly available. It achieves 93.9% on SWE-bench Verified and demonstrates …
MTP (Multi-Token Prediction) is an inference acceleration technique that lets a lightweight drafter model predict several future tokens simultaneously, which a larger target model then verifies in a s…
Natural Language Autoencoders (NLAs) are an unsupervised interpretability technique that converts a language model's internal activations into plain-text explanations. An activation verbalizer transla…
ProgramBench is a software-engineering benchmark that tests whether AI agents can reconstruct a complete, working codebase from only a compiled binary and its documentation — no source code, no decomp…
Zaya1-8B is an open-weight mixture-of-experts reasoning model from [Zyphra](https://www.zyphra.com/post/zaya1-8b) that activates only 760 million of its 8.4 billion parameters per forward pass, delive…
Airbyte Agents is a context layer that gives AI agents unified, search-optimized access to an organization's operational data before the agent ever runs — solving what the company calls the core produ…
Inverse laws are three human-conduct principles that mirror-invert Asimov's Three Laws of Robotics. Where Asimov's laws constrain robot behavior toward human safety, these invert the target: they cons…
SubQ is the brand name of the first large language model built on a fully sub-quadratic architecture, where attention compute scales linearly rather than quadratically with context length. The model i…
Token-maxxing is the practice of maximizing AI token consumption as a proxy for productivity — competing on internal leaderboards, running autonomous agents around the clock, and treating spend as sta…
AI psychosis names two converging phenomena: the clinical pattern where vulnerable users develop or deepen delusions through heavy chatbot use, and the builder-culture slang where developers describe …
A coding plan is a fixed-price monthly subscription that gives developers quota-capped access to frontier AI models via their existing coding tools — Cursor, Claude Code, Cline — replacing per-token b…
Virgo Network is Google's custom-built megascale AI data center fabric, announced April 22, 2026 at Google Cloud Next '26. It replaces the multi-tier Jupiter Ethernet design with a flat, two-layer non…
AI Supply refers to the physical and logistical capacity needed to deliver AI compute at scale — GPUs, high-bandwidth memory, advanced packaging, power infrastructure, and data-center buildout. As tok…
Flue is an open-source TypeScript framework for building autonomous agents, positioned as "The Agent Harness Framework" — not another AI SDK, but infrastructure that wraps models with a programmable h…
Granite 4.1 is IBM's latest open-source language model family, released April 29, 2026, in dense 3B, 8B, and 30B parameter sizes under an Apache 2.0 license. It targets enterprise workloads emphasizin…
Visual Primitives are coordinate-based reasoning anchors — points and bounding boxes — embedded directly into an AI model's chain-of-thought rather than output only as final answers. The term names a …
GLM-4.7 is an open-weight large language model from Z.ai (formerly Zhipu AI) designed for agentic coding, multi-step reasoning, and production tool-use workflows. The 358B-parameter Mixture-of-Experts…
GLM-5 is Zhipu AI's (Z.ai) flagship open-weight large language model, a 744-billion-parameter Mixture-of-Experts system built for agentic coding and long-horizon tool use. Released under the MIT licen…
hermes.md is the project-specific configuration file used by [Hermes Agent](https://github.com/NousResearch/hermes-agent) (NousResearch), the 126k-star AI coding agent. Stored at the git root as `HERM…
Laguna XS.2 is Poolside's first open-weight model: a 33B-total / 3B-active Mixture-of-Experts (MoE) coding model released under Apache 2.0 on April 28, 2026. Designed for agentic coding and long-horiz…
Long-running agents are AI agents designed to sustain work across multiple context windows, persisting state through structured artifacts — progress files, git commits, feature specs — so each new ses…
Value Accuracy measures the fraction of JSON leaf values that exactly match ground truth — distinct from JSON pass rate, which only checks schema structure. The gap is large: schema compliance routine…
Wanman is an open-source agent matrix runtime that orchestrates multiple specialized AI agents — CEO, dev, devops, marketing, feedback — as isolated CLI subprocesses on your local machine. The name de…
A workspace agent is an AI agent scoped to a shared organizational environment — running persistently in the cloud, accessing team-connected tools, and acting autonomously on behalf of the whole team.…
Agent click refers to the capability of an AI agent to click, type, scroll, and read native desktop applications in the background without stealing the user’s cursor, window focus, or active Space. Th…
CUA (Computer-Use Agent) is the open-source infrastructure project by [trycua](https://github.com/trycua/cua) that provides sandboxes, SDKs, and benchmarks for building and evaluating AI agents that c…
Gemma 4 is Google DeepMind's fourth-generation family of open-weight multimodal models, released April 2, 2026 under Apache 2.0. Four sizes span phones to data centers: E2B, E4B, 26B Mixture-of-Expert…
Talkie (full name: talkie-1930) is a 13B open-weight language model trained exclusively on 260 billion tokens of English text published before 1931. Its hard knowledge cutoff at December 31, 1930 make…
Xiaomi MiMo Orbit is Xiaomi's open developer ecosystem program, pairing a 100-trillion-token grant for AI builders with an Agent Ecosystem co-construction track for framework teams — Xiaomi's bet that…
DeepSeek V4 Pro is the premium tier of DeepSeek's V4 series: a 1.6-trillion-parameter, 49-billion-active Mixture-of-Experts model with a 1-million-token context window. It is the largest open-weight m…
Shared Agents is the emerging paradigm for AI agents built once and deployed as an organizational resource — accessible, reusable, and improvable by an entire team rather than a single user. The shift…
A Stop hook is the [Claude Code lifecycle event](https://code.claude.com/docs/en/hooks) that fires every time the agent finishes a response turn. Shell commands, HTTP endpoints, or LLM prompts configu…
The tool layer is the discrete architectural component of an AI agent harness that registers, validates, and gates every function the model can call. It sits between the orchestration loop above and t…
Wiki Layer is an architectural pattern for multi-agent systems: a shared, git-native markdown store that all agents on a team can read from and write to, distinct from each agent's private notebook. I…
DeepSeek V4 is a series of open-weight Mixture-of-Experts language models from DeepSeek that bring one-million-token context to frontier-class performance at a fraction of closed-source pricing. The s…
Gemini 3.1 Pro is Google DeepMind's flagship reasoning model, released [February 19, 2026](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/). It introduces a "th…
GPT-5.5 is OpenAI's frontier language model released on April 23, 2026 — the first fully retrained base model since GPT-4.5, with every GPT-5.x release between them being a post-training iteration. De…
Privacy Filter is an open-weight, on-device model for detecting and redacting personally identifiable information (PII) from unstructured text. It runs locally — no data leaves the machine — making it…
SuperHQ is an open-source macOS app that runs AI coding agents — Claude Code, OpenAI Codex, and Pi — inside isolated microVM sandboxes instead of directly on your machine. A local auth gateway keeps A…
Agent Teams is the [Claude Code](https://code.claude.com/docs/en/agent-teams) feature that lets multiple Claude instances coordinate on a shared codebase in parallel. One session is the team lead; eac…
AI-native describes a product, company, or workflow designed from the ground up with artificial intelligence as a foundational architectural element — not bolted on after launch. If you removed the AI…
Broccoli is an open-source coding agent that converts [Linear](https://linear.app) project-management tickets into shipped pull requests, running each task inside an isolated Google Cloud sandbox. Bui…
Cloud coding agents are AI software engineering systems that execute development tasks inside remote, sandboxed cloud environments rather than on a developer's local machine. They plan, write code, ru…
GLM-5.1 is Z.ai's 754-billion-parameter open-weight large language model, purpose-built for agentic engineering and long-horizon coding tasks. It is a post-training upgrade to GLM-5 sharing the same M…
LLM-as-a-Judge is an evaluation pattern where a large language model scores or ranks outputs from another AI system, replacing expensive human reviewers with an automated judge that applies a natural-…
Manus is a general-purpose autonomous AI agent that accepts plain-language goals and executes multi-step tasks — web research, code generation, data analysis, file creation — without continuous user i…
ML Intern is an open-source autonomous AI agent that runs the full machine-learning research loop without human supervision. It reads papers on arXiv, sources and reformats datasets from Hugging Face …
Preflight is a free validator that probes a live MCP server endpoint and checks every requirement — OAuth 2.1 + PKCE flow, CORS headers, protocol handshake, tool execution, domain verification — befor…
Roo Code is an open-source VS Code extension that turns the editor into a multi-agent coding workspace, letting developers orchestrate specialized AI agents across Architect, Code, Debug, Ask, and Cus…
Seedance is a text-to-video AI model family developed by ByteDance's Seed research team that generates cinematic, multi-shot video from text and image inputs. It couples video generation with native s…
Workspace Agents is OpenAI's enterprise feature that lets teams build shared AI agents — powered by Codex — capable of running multi-step workflows autonomously across connected business tools. Unlike…
Code Search MCP is a category of [Model Context Protocol](https://www.anthropic.com/news/model-context-protocol) servers that give AI coding agents semantic and structural search over local codebases.…
Daemons are self-initiated AI background processes that handle ongoing maintenance work in software repositories — watching for events, detecting drift, and executing routines without prompts. Unlike …
Deep Research is an agentic AI capability that autonomously browses the web, synthesizes hundreds of sources, and produces a cited analyst-level report from a single prompt. The agent iterates — searc…
An OpenClaw Skill is a self-contained capability package — a directory with a [`SKILL.md`](https://docs.openclaw.ai/tools/skills) file — that teaches the OpenClaw agent how to interact with an externa…
Agentic frameworks are software toolkits that wire a language model into a running agent — orchestrating the loop, tool calls, memory, and multi-agent coordination so developers don't build that plumb…
An architecture diagram generator is an AI tool that turns a prose description, a code repository, or infrastructure-as-code (Terraform, CloudFormation) into a rendered system architecture diagram — b…
Browser Harness is a 592-line Python project from [browser-use](https://github.com/browser-use/browser-use) that gives an LLM direct, unmediated control of Chrome via the [DevTools Protocol](https://b…
[Claude Opus 4.6](https://www.anthropic.com/news/claude-opus-4-6) is Anthropic's flagship LLM released February 5, 2026. It was the first Opus-class model with a 1M-token context window (beta at launc…
[Cursor CLI](https://cursor.com/cli) is [Cursor](https://cursor.com)'s headless command-line agent — the same Cursor Agent that drives the IDE, callable from any terminal or CI runner without the edit…
**GPT Image 2** is the anticipated second-generation release of OpenAI's [gpt-image](/term/gpt-image) model family — the natively multimodal image engine that powers ChatGPT's image tool and the `gpt-…
**gpt-image** is OpenAI's natively multimodal image generation and editing API family — the same engine that powers ChatGPT's image tool, exposed as model IDs `gpt-image-1`, `gpt-image-1-mini`, and `g…
Information Operating System is the self-applied tagline for [GalaxyBrain](https://galaxybrain.com/), a local-first knowledge workspace where every page carries variables, formulas, and live cross-ref…
[Kimi K2.6](https://www.kimi.com/blog/kimi-k2-6) is Moonshot AI's April 20, 2026 open-weight flagship — a 1T-parameter Mixture-of-Experts model (32B active, 384 experts, 256K context, native multimoda…
LLM wiki is a pattern where an AI agent incrementally compiles a folder of plain-markdown entity pages from your raw sources, then answers questions against that structured wiki instead of re-retrievi…
Mediator AI is the emerging category of LLM-driven tools that act as an impartial third party to resolve disputes between two or more people — elicit each side's preferences privately, propose candida…
OpenAI Agents SDK is a lightweight open-source framework for building multi-agent workflows on top of OpenAI models. It ships four primitives — Agents, Handoffs, Guardrails, and Sessions — plus built-…
QMD is a local-first command-line search engine for Markdown notes, docs, and knowledge bases, built by Shopify CEO [Tobi Lütke](https://github.com/tobi/qmd). It combines BM25 full-text search, vector…
RLMs (Recursive Language Models) are an inference strategy where an LLM treats its prompt as an object inside a Python REPL, then recursively calls sub-LLMs over chunks of it instead of stuffing every…
The UserPromptSubmit hook is a [Claude Code](https://code.claude.com/docs/en/hooks) lifecycle event that fires the moment a user hits enter on a prompt, before Claude processes it. A shell command con…
Vibe Island is a native macOS notch utility that turns the Dynamic Island area into a status-and-approval HUD for AI coding agents. It watches Claude Code, Codex, Cursor, Gemini CLI and seven other CL…
"Agent traps" is the shorthand English phrase that maps one-to-one to AI Agent Traps, the taxonomy Google DeepMind published on March 27, 2026 for adversarial web content that hijacks autonomous AI ag…
Agentic design is the discipline of architecting software systems where one or more AI agents reason, plan, and act autonomously over tools and memory. It names the pattern layer above a single LLM ca…
AI agent traps are adversarial web content designed to manipulate, hijack, or weaponize autonomous AI agents against the users they serve. The phrase names a category, not a product: six attack famili…
An AI stack is the layered set of components that turns raw compute into a shippable AI product — chips and models at the bottom, data and orchestration in the middle, the application on top. Menlo Ve…
An AI workspace is a persistent, project-scoped container where chats, files, tools, and memory live together — replacing the stateless single-thread chatbot. Notion markets itself as "[the AI workspa…
Apple Intelligence is Apple's on-device AI brand announced at [WWDC 2024](https://www.apple.com/newsroom/2024/06/introducing-apple-intelligence-for-iphone-ipad-and-mac/) and shipped across iOS 26, iPa…
AutoResearch is an agent loop in which an LLM autonomously edits a single training file, runs a fixed 5-minute experiment, checks whether a chosen metric improved, and keeps or reverts the change — re…
Browser Use is an open-source Python library ([browser-use/browser-use](https://github.com/browser-use/browser-use)) that lets an LLM drive a real Chrome instance to complete web tasks — click, scroll…
busybee is an open-source FIFO build queue for multi-agent developer workflows. It wraps [pueue](https://github.com/Nukesor/pueue) and gates heavy commands — cmake, cargo, ninja, large test suites — s…
Dev Agents is a loose umbrella label for AI agents that write, review, and ship code on a developer's behalf — a near-synonym of the more established [coding agents](/term/coding-agents) category. In …
DGX Spark is NVIDIA's $3,000-$4,000 desktop AI supercomputer — a 1.2 kg box built around the [GB10 Grace Blackwell Superchip](https://www.nvidia.com/en-us/products/workstations/dgx-spark/) with 128 GB…
ElevenLabs is a voice AI company that builds text-to-speech, speech-to-text, voice cloning, and voice-agent infrastructure used by developers, publishers, and enterprises. Founded in 2022 by ex-Google…
Feral (stylized FeralHq) is a new agentic content-generation product that writes humor-first social posts for brand accounts. It crawls a company's website to build a brand profile, then runs a genera…
GRPO (Group Relative Policy Optimization) is a reinforcement-learning algorithm that teaches language models to reason by sampling multiple answers per question and scoring each answer against the gro…
Infrawise is an AI-agent Azure cost-optimization tool that plugs into a customer's cloud environment through a read-only allowlist of Azure APIs and returns prioritized rightsizing, idle-resource, and…
Kiro is AWS's agentic AI IDE, built on Code OSS and pitched as a fix for "the chaos of vibe coding." Its headline bet: specs first, code second. Every task starts as a generated requirements + design …
An [MCP tool](https://modelcontextprotocol.io/docs/concepts/tools) is one executable function exposed by an [MCP server](https://modelcontextprotocol.io/) to a language model over JSON-RPC, defined by…
"MCPs" is the informal plural of [MCP](https://www.anthropic.com/news/model-context-protocol) — the Model Context Protocol Anthropic released on November 25, 2024. When developers say "I kept adding M…
MLX is Apple's open-source array framework for machine learning on Apple Silicon. Its API mirrors NumPy and PyTorch, but the whole runtime is built on Metal and the unified memory architecture, so ope…
Personal AI is the category of AI systems built around one individual — their memory, voice, preferences, and recall — rather than a generic assistant shared by millions. Incarnations range from dedic…
"Spy AI" is a shared-string bucket, not one product: the phrase collides across AI trading bots for the SPDR S&P 500 ETF (ticker SPY), a Virtuals token called [SPY AI AGENT](https://www.coingecko.com/…
Vault Context is the practice of feeding a local Markdown vault — its files, folder layout, frontmatter, tags, and retrieved chunks — to an LLM so the model answers grounded in your own notes instead …
AEO (Answer Engine Optimization) is the practice of structuring content so that AI answer engines — ChatGPT, Perplexity, Google AI Overviews, Claude — pick your page as the extractable answer to a use…
HyperFrames is an open-source video rendering framework from HeyGen that lets AI agents compose videos by writing HTML, CSS, and JavaScript, then outputs MP4, MOV, or WebM via Puppeteer + FFmpeg. The …
Intelligence Age is OpenAI's preferred epochal framing for the AI era — the successor to the Stone, Agricultural, and Industrial Ages in a narrative where compute, energy, and frontier models drive th…
Slop Cop is an automated prose linter that flags the rhetorical and structural tells of LLM-generated writing — em-dash pivots, hedge stacks, throat-clearing intros, three-item lists, filler intensifi…
AI Agent Identity is the emerging set of protocols and file formats that let an autonomous agent prove who it is, what it's authorized to do on whose behalf, and what values it embodies. Two layers: c…
AI CapEx is the industry shorthand for capital expenditure on AI infrastructure — the land, power, GPUs, servers, and data centers that hyperscalers are building to train and serve large models. It ha…
GEO (Generative Engine Optimization) is the practice of structuring content so that large language models — ChatGPT, Google Gemini, Perplexity, Claude — cite, quote, or recommend your brand inside the…
Prism is a unified pull-request inbox that consolidates GitHub and GitLab review queues into one prioritized feed. Its pitch — ["keep up with code review, even as AI speeds up the rest"](https://prism…
Claude Design is an Anthropic Labs product that turns a conversation with Claude into polished visual artifacts — prototypes, slides, one-pagers, pitch decks, and marketing collateral — rendered as li…
[ClawHub](https://clawhub.ai/) is the public skill registry and marketplace for [OpenClaw](https://github.com/openclaw/openclaw), the self-hosted personal AI agent. Developers publish text-based `SKIL…
Context engineering is the discipline of curating every token that enters an LLM's context window — system prompt, tools, retrieved data, conversation history, memory, files — so the model can plausib…
GPT-5.4 is OpenAI's March 2026 frontier language model that unifies the Codex and GPT product lines into a single system, adding native computer-use capabilities, a 1-million-token context window, and…
Qwen3.6 is Alibaba's Qwen team's next-generation LLM line, positioned around "real-world agents." It spans two tiers: the closed [Qwen3.6-Plus](https://qwen.ai/blog?id=qwen3.6) (released April 2, 2026…
Qwen3 is Alibaba's third-generation open-weight foundation model family, launched [April 28, 2025](https://qwenlm.github.io/blog/qwen3/) under Apache 2.0. It introduced a hybrid 'Thinking' / 'Non-Thin…
Tokenmaxxing is the practice — and increasingly the critique — of treating AI token consumption as a productivity metric. Engineers ship more AI-generated code, climb internal token-burn leaderboards,…
An agent harness is the middleware between a large language model and the real world — code that runs the agent loop, calls tools, manages memory, enforces guardrails, and recovers from errors. Practi…
An agent loop is the control-flow pattern at the center of every autonomous LLM agent: the model observes its context, reasons about what to do, calls a tool, reads the result, and loops — until it pr…
Agentic AI names a class of AI systems that autonomously plan, decide, and take actions to meet user-defined goals — not single-shot chatbot answers, but a goal-driven loop that calls tools, updates s…
Agentic coding is the software-development pattern where an autonomous AI agent plans, writes, tests, and iterates on code against a feedback loop — running its own builds, reading its own errors, and…
Coding Agents is the category name for AI developer tools that act on code autonomously — reading a repo, planning a change, editing files, running tests, and opening pull requests — rather than sugge…
A [context window](https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents) is the span of tokens an LLM reads and reasons over in a single forward pass. In 2025-2026 the phr…
[Gemini 3.1 Flash TTS](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-flash-tts/) is Google DeepMind's text-to-speech model that generates expressive speech in 70+ …
Gemini 3.1 Flash is Google's mid-2026 speed-tier refresh of the Gemini model family — a family-of-variants brand rather than a single model. Developers use "3.1 Flash" as the umbrella query; Google sh…
LM Studio is a desktop GUI — Windows, macOS, Linux — for discovering, downloading, and running open-source large language models entirely on your own machine. It wraps llama.cpp, MLX, and a visual Hug…
Managed Agents is an infrastructure paradigm where cloud platforms host, orchestrate, and operate AI agents as a service. Instead of developers hand-building agent loops — memory, tool routing, state,…
An [MCP server](https://modelcontextprotocol.io/) is a small, standalone program that exposes one capability — a database, a filesystem, a security scanner, a trading API — to AI agents over the [Mode…
Nano Banana is the codename, and now the public brand, for [Google DeepMind’s Gemini image-generation family](https://deepmind.google/models/gemini-image/flash/). The original Nano Banana shipped as ‘…
Parallel Agents is the pattern of running multiple AI coding sessions at the same time against isolated copies of a codebase, with a human orchestrator switching between them as each returns results. …
Qwen (通义千问) is [Alibaba Cloud's](https://qwenlm.github.io/) open-weight large-language-model family, shipped by the Tongyi Lab since August 2023. It spans text, code, vision, audio, image, embedding, …
AGENTS.md is an open, vendor-neutral markdown file placed at the root of a repository that tells AI coding agents (Claude Code, Codex CLI, Cursor, Copilot, Jules, Amp, Factory, Windsurf) how to work w…
AI slop is a pejorative for generative-AI content — text, images, video, pull-requests — that is technically fluent but intellectually empty, mass-produced for clicks or compliance rather than communi…
[Claude Agent SDK](https://code.claude.com/docs/en/agent-sdk/overview) is Anthropic's programmatic toolkit for building AI agents on Claude. It exposes the same tool-loop, context management, subagent…
[Claude Code](https://claude.com/product/claude-code) is Anthropic's official command-line coding agent — a terminal tool that reads your codebase, edits files, runs commands, opens PRs, and drives lo…
[Claude Mythos](https://red.anthropic.com/2026/mythos-preview/) is an unreleased frontier LLM from Anthropic, publicly previewed on April 7, 2026. Positioned as general-purpose, the preview centres on…
[Claude Opus 4.7](https://www.anthropic.com/news/claude-opus-4-7) is Anthropic's flagship LLM, released April 16, 2026. It narrowly retakes the lead over GPT-5.4 and Gemini 3.1 Pro on agentic coding b…
Context rot is the measurable degradation in large-language-model output quality as input length grows, even when the prompt stays well under the advertised context window. Models don't process the 10…
[Model Context Protocol](https://modelcontextprotocol.io/) (MCP) is an open, JSON-RPC-2.0-based standard that defines how AI applications talk to external tools, data, and systems. The spec defines a …
[OpenAI Codex CLI](https://developers.openai.com/codex/cli) is OpenAI's official terminal-resident coding agent: it reads a repo, edits files, runs commands, invokes subagents, and drives long tasks v…
[OpenClaw](https://openclaw.ai/) is an open-source self-hosted personal AI agent: a long-running runtime that connects to any LLM (Claude, GPT, DeepSeek, Kimi, Gemini) and drives itself from messaging…
OpenCode is an [open-source AI coding agent](https://opencode.ai/) that runs in the terminal and connects to any model provider — Claude, GPT, Gemini, local LLMs, 75+ providers via Models.dev. It’s th…
[Hermes Agent](https://hermes-agent.nousresearch.com/) is an MIT-licensed, self-improving AI agent from Nous Research designed to live on a server — local, VPS, or serverless — and accumulate capabili…