The AI terms that
actually matter.
Clear, technical definitions of 146 key concepts — written by AI engineers, not marketers.
OmniVoice
A massively multilingual open-source TTS model by Xiaomi supporting 646 languages with zero-shot voice cloning and voice design, running 40× faster than real-time.
Read definitionOpenClaw
OpenClaw is an open-source, local-first autonomous AI agent that runs as a persistent daemon on your own hardware, connecting to any LLM backend and operated through messaging platforms like WhatsApp, Telegram, or Discord.
Read definitionOrpheus TTS
A Llama 3-based open-source TTS model by Canopy Labs with human-level expressiveness, inline emotion tag control, zero-shot voice cloning, and real-time streaming — the highest-quality open voice model for conversational AI.
Read definitionPagedAttention
PagedAttention is the memory-management technique behind vLLM and most modern LLM serving engines: it stores each request's KV cache in small fixed-size blocks scattered anywhere in GPU memory, with a per-request block table mapping logical token positions to physical blocks, exactly like an operating system's virtual memory. It removed the fragmentation that used to waste most of the KV cache, which is why one GPU can now serve several times more concurrent requests.
Read definitionPhysical AI
Physical AI refers to embodied artificial intelligence systems designed to sense, comprehend, and interact directly with the physical world, bridging the gap between digital cognition and real-world action.
Read definitionPi (Coding Agent Harness)
Pi is a minimal, open-source (MIT) coding-agent harness created by Mario Zechner (earendil-works). Its thesis: a capable coding agent needs only four tools — read, write, edit, bash — and a tiny system prompt, with everything else added as hot-reloadable TypeScript extensions. Despite shipping without MCP, sub-agents, or plan mode, Pi ranked #2 on Terminal-Bench with Claude Opus.
Read definitionPrefill-Decode Disaggregation (PD Disaggregation)
Prefill-decode disaggregation is an LLM serving architecture that runs the two phases of inference, prefill (reading the prompt) and decode (generating tokens one at a time), on separate GPU pools and ships the KV cache between them. Because prefill is compute-bound and decode is memory-bandwidth-bound, splitting them removes the interference that makes colocated serving stall streaming responses, and lets each pool be sized and tuned independently.
Read definitionPrefix Caching and Position-Independent Caching (PIC)
Prefix caching reuses the KV cache computed for the start of a prompt (a system prompt, tool definitions, a long document) across later requests, so the model skips prefill for text it has already processed; it only works when everything before the reused text is identical. Position-independent caching (PIC) relaxes that rule so cached chunks can be reused even when they appear at a different position or after different text, which is what RAG and agent workloads actually produce.
Read definitionPrompt Engineering
Prompt engineering is the practice of designing and refining the text inputs given to an AI model to reliably produce accurate, useful, and well-formatted outputs — without changing the model's weights.
Read definitionPrompt Injection
Prompt Injection is a critical vulnerability in Large Language Models (LLMs) where malicious, untrusted instructions are embedded into user inputs or external data to manipulate the model into ignoring its safety guidelines and executing unintended actions.
Read definitionProsody
Prosody is the melody, rhythm, and stress pattern of speech, carried by pitch (F0), duration, and energy, which turns flat, monotone synthetic audio into speech that sounds like it means something.
Read definitionQuery Fan-Out
Query fan-out is the technique of expanding a single user query into many related sub-queries, running them concurrently against one or more sources, and fusing the results into one answer. It powers Google's AI Mode and the RAG-Fusion pattern, and it changes what it means to rank for a search.
Read definitionStay ahead of the curve
Weekly newsletter on agentic AI, LLMs, and what we're building at Superteams — straight to your inbox.
Ready to ship AI in production?
We deploy fractional AI teams that deliver production-grade systems in 30–90 days. No fluff, no obligation.