Know the language. Build with clarity.

414 defined terms · 16 topic areas · 50 essentials

The complete reference

414 matching terms.

• A2A Protocol · Protocols & interoperability

  Agent2Agent: a protocol for communication between independently implemented agentic applications.

• ACP Protocol · Protocols & interoperability

  Agent Client Protocol: standard communication between coding agents and editor or IDE clients.

• Adversarial agent Concept · Loops, critics & adversaries

  An agent assigned to oppose, challenge, or attack a target within a defined setting.

• Adversarial loop Pattern · Loops, critics & adversaries

  An informal pattern of challenge, revision, and rechecking; not one standardized algorithm.

• Agent Concept · Agents & architecture

  A system that selects actions, observes their results, and continues toward an objective.

• Agent loop Pattern · Loops, critics & adversaries

  Repeatedly assemble context, request an action, execute tools, observe results, and continue or stop.

• Bounded autonomy Concept · Agents & architecture

  Independent action within explicit permissions, budgets, scope, and stop conditions.

• Compaction Concept · Context & memory

  Replace detailed history with a smaller representation that preserves task-relevant information.

• Context engineering Concept · Context & memory

  Select, structure, and update the information available to the model throughout execution.

• Cost per successful task Concept · Evals & observability

  Total execution cost divided by successful outcomes, including the cost of unsuccessful attempts.

• Critic Concept · Loops, critics & adversaries

  A component that assesses a candidate and identifies weaknesses or potential improvements.

• Durable execution Concept · Security & reliability

  Recover and continue a workflow across interruptions using persisted state or event history.

• Embedding Concept · Retrieval & knowledge

  A learned vector representation used for similarity and other downstream tasks.

• Eval Concept · Evals & observability

  A systematic test of a model or application's behavior against defined criteria.

• Generative UI Concept · Generative UI & voice

  Use model output to help determine which interface components and content a user sees.

• Grounding Concept · Retrieval & knowledge

  Tie generated claims or actions to relevant evidence and observations.

• Handoff Pattern · Multi-agent coordination

  Transfer responsibility or conversational control to another agent.

• Harness Concept · Agents & architecture

  The surrounding execution system: model calls, tools, state, context, permissions, and stopping rules.

• Harness engineering Concept · Agents & architecture

  Improving the environment and controls around an agent, rather than only its prompt or model.

• Hybrid search Concept · Retrieval & knowledge

  Combine complementary retrieval methods, commonly lexical and dense vector search.

• Idempotency Concept · Security & reliability

  Repeated execution of the same operation has the same intended effect as one execution.

• Inference-time compute Concept · Models, reasoning & training

  Computation spent solving a request rather than training the model.

• KV cache Concept · Inference & performance

  Stored attention keys and values reused to avoid recomputing prior-token attention state.

• Least privilege Concept · Security & reliability

  Grant only the access and capabilities necessary for a specific task.

• LLM-as-a-judge Concept · Evals & observability

  Use a model to assess outputs or execution trajectories against a rubric.

• LoRA Concept · Models, reasoning & training

  Low-rank adaptation: learn low-rank parameter updates instead of updating all model weights.

• MCP Protocol · Protocols & interoperability

  Model Context Protocol: a standard interface for exposing tools, resources, and prompts to AI applications.

• Model routing Concept · Inference & performance

  Choose a model based on task needs, constraints, policy, or observed difficulty.

• Plan Concept · Goals, specs & plans

  A proposed strategy and sequence for executing the work.

• Programmatic tool calling Concept · Tools & output contracts

  Use model-generated code to orchestrate tool calls and process intermediate outputs.

• Progressive disclosure Concept · Context & memory

  Expose concise metadata first and fuller instructions or resources only when relevant.

• Prompt caching Concept · Inference & performance

  A provider's mechanism for reusing eligible prompt processing; exact behavior and billing are platform-specific.

• Prompt injection Concept · Security & reliability

  Untrusted content attempts to redirect a model away from the application's intended instructions.

• RAG Pattern · Retrieval & knowledge

  Retrieval-augmented generation: fetch external evidence and use it to support a generated response.

• Repair loop Pattern · Loops, critics & adversaries

  Modify a candidate or strategy in response to diagnosed failure, then try again.

• Replanning Concept · Loops, critics & adversaries

  Revise the execution plan after new information, failures, or changed constraints.

• Reranker Concept · Retrieval & knowledge

  A component that re-scores an initial candidate set for relevance.

• SFT Concept · Models, reasoning & training

  Supervised fine-tuning: train on examples of desired inputs and outputs.

• Skill Concept · Coding-agent internals

  A reusable procedure or knowledge package that an agent can load when relevant.

• Spec-driven development Pattern · Goals, specs & plans

  A workflow in which explicit specifications guide design, implementation, and verification.

• Specification Concept · Goals, specs & plans

  A precise description of required behavior, interfaces, or properties.

• Stop condition Concept · Loops, critics & adversaries

  An explicit rule for completion, failure, budget exhaustion, cancellation, or escalation.

• Structured outputs Concept · Tools & output contracts

  Model responses constrained to a supported output schema.

• Subagent Concept · Multi-agent coordination

  A delegated agent with its own task scope and often separate context or tool access.

• Task success rate Concept · Evals & observability

  The fraction of evaluated tasks meeting the specified completion criteria.

• Tool calling Concept · Tools & output contracts

  The model proposes a named operation and arguments; the runtime validates and executes the operation.

• Tool discovery Concept · Tools & output contracts

  Find relevant tool capabilities at runtime instead of preloading every definition.

• Trace Concept · Evals & observability

  A structured record of model calls, tool actions, timing, errors, and related execution events.

• TTFT Concept · Inference & performance

  Time to first token: delay from a defined request start until the first generated token arrives.

• Verifier Concept · Loops, critics & adversaries

  A component that checks claims or properties against evidence and specified conditions.

• A2UI Protocol · Generative UI & voice

  A declarative format for agent-generated interfaces rendered by compatible clients.

• Ablation Concept · Evals & observability

  Remove or change a component to estimate its contribution to system performance.

• Acceptance criteria Concept · Goals, specs & plans

  Concrete conditions used to accept or reject a feature or deliverable.

• Active parameters Concept · Models, reasoning & training

  The subset of model parameters used for a given token or computation path.

• Actor-critic pattern Pattern · Loops, critics & adversaries

  A producer and assessor arrangement; distinguish this loose agent pattern from actor-critic reinforcement learning.

• ADR Concept · Goals, specs & plans

  An architecture decision record documenting a decision, context, alternatives, and consequences.

• Adversarial Concept · Loops, critics & adversaries

  Designed to challenge, exploit, or oppose a target system or candidate outcome.

• Adversary Concept · Loops, critics & adversaries

  An actor or process whose objective conflicts with the target system's objective.

• AG-UI Protocol · Generative UI & voice

  Agent User Interaction Protocol: event-oriented communication between agent execution and user interfaces.

• Agent Card Concept · Protocols & interoperability

  An A2A capability description used to discover how to interact with an agent.

• Agent framework Concept · Agents & architecture

  A developer library offering abstractions for agents, tools, state, or workflows.

• Agent runtime Concept · Agents & architecture

  The execution machinery that processes model calls, tool requests, state transitions, and recovery.

• Agent team Concept · Multi-agent coordination

  A collection of agents with assigned roles and a defined way to coordinate.

• Agent-as-a-tool Pattern · Multi-agent coordination

  A parent invokes a subordinate agent and receives its result while retaining overall control.

• Agent-computer interface Concept · Tools & output contracts

  The names, arguments, outputs, errors, and affordances through which an agent operates tools.

• Agentic AI Concept · Agents & architecture

  A broad label for systems where models select or sequence actions toward an objective; the label does not specify autonomy, tools, or controls.

• Agentic engineering Concept · Agents & architecture

  Software engineering with agents while retaining explicit requirements, verification, and control.

• Agentic RAG Pattern · Retrieval & knowledge

  Let an agent decide what to retrieve, whether evidence is sufficient, and when to search again.

• Agentic workflow Concept · Agents & architecture

  A workflow containing model-directed decisions; it can still have deterministic stages and approvals.

• AgentOps Concept · Agents & architecture

  Operational practices focused on agent trajectories, actions, permissions, reliability, and cost.

• AGENTS.md File convention · Coding-agent internals

  Repository guidance for coding agents, such as conventions, architecture, test commands, and constraints.

• AI engineering Concept · Agents & architecture

  Engineering applications around models: interfaces, data, tools, evaluations, deployment, and operations.

• AI gateway Concept · Agents & architecture

  A service that centralizes model routing, policy, observability, and provider access for AI applications.

• AI-native Concept · Agents & architecture

  A broad product label implying AI shapes the core workflow rather than being an incidental feature.

• Amazon Bedrock AgentCore Tool · Tools & ecosystem

  AWS services for operating and integrating agents, including managed runtime capabilities.

• Ambient agent Concept · Loops, critics & adversaries

  An informal label for an agent that monitors relevant events rather than requiring a prompt for every action.

• ANN Concept · Retrieval & knowledge

  Approximate nearest-neighbor search: trade some exactness for efficient similarity lookup.

• Approval gate Concept · Security & reliability

  Require authorization before a sensitive action proceeds.

• Artifact Concept · Coding-agent internals

  A concrete output such as a patch, screenshot, report, test log, or build result.

• ASR / STT Concept · Generative UI & voice

  Automatic speech recognition or speech-to-text: transcribe spoken audio into text.

• AST Concept · Coding-agent internals

  Abstract syntax tree: a structural representation of parsed source code.

• Attention Concept · Models, reasoning & training

  A learned mechanism for combining information from different input positions or representations.

• Audit log Concept · Security & reliability

  A durable record of actions and relevant identity, authorization, and outcome information.

• AutoGen Tool · Tools & ecosystem

  A Microsoft-originated framework for agent applications and multi-agent interaction.

• Backpressure Concept · Loops, critics & adversaries

  Feedback or limits that slow or reject progress; in coding-agent discussion, often tests, types, and lint failures.

• Barge-in Concept · Generative UI & voice

  Allow a user to interrupt an agent's speech and have the system respond appropriately.

• Base model Concept · Models, reasoning & training

  A pretrained model before a particular instruction or preference adaptation stage.

• Batch inference Concept · Inference & performance

  Process a collection of requests as an offline or scheduled workload.

• Benchmark contamination Concept · Evals & observability

  Evaluation material leaks into training or optimization, weakening the validity of the evaluation.

• Best-of-N Pattern · Models, reasoning & training

  Generate multiple candidates and select one using a scoring method.

• Blackboard architecture Pattern · Multi-agent coordination

  Coordinate independent contributors through a shared workspace of facts, tasks, and partial solutions.

• Blind review Concept · Loops, critics & adversaries

  Assess an artifact without revealing irrelevant producer identity or prior assessments that could bias judgment.

• BM25 Concept · Retrieval & knowledge

  A lexical relevance-ranking method based on term occurrence and document statistics.

• BMAD Method Tool · Tools & ecosystem

  An AI-assisted development methodology organized around roles, planning artifacts, and workflows.

• Browser automation Concept · Coding-agent internals

  Programmatic interaction with a browser to inspect or operate a running application.

• Budget enforcement Concept · Security & reliability

  Mechanically limit execution by time, tokens, money, iterations, or allowed actions.

• Cancellation Concept · Security & reliability

  A request to stop work; already-completed external side effects may require separate compensation.

• Capability negotiation Concept · Protocols & interoperability

  Participants establish which supported protocol features they can use together.

• Chain of thought Concept · Models, reasoning & training

  Intermediate reasoning expressed in steps; visible explanations are not guaranteed faithful traces of internal computation.

• Checkpointing Concept · Security & reliability

  Persist execution state so work can resume from a known boundary.

• Chunk overlap Concept · Retrieval & knowledge

  Repeat boundary content across adjacent chunks to preserve local context.

• Chunking Concept · Retrieval & knowledge

  Split source material into units for indexing, retrieval, or processing.

• Circuit breaker Pattern · Security & reliability

  Temporarily stop calling a failing dependency to limit repeated failures and allow recovery.

• Claude Agent SDK Tool · Tools & ecosystem

  Programmatic tooling for building agent applications using capabilities associated with Claude Code.

• Claude Code Tool · Tools & ecosystem

  Anthropic's coding-agent environment with repository tools and project instructions.

• CLAUDE.md File convention · Coding-agent internals

  A project instruction file used by Claude Code; loading and precedence are product-specific.

• Codex Tool · Tools & ecosystem

  OpenAI coding-agent tooling; distinguish the application and harness from the specific model used.

• Compensating transaction Concept · Security & reliability

  An explicit business operation that counteracts a completed action when rollback is unavailable.

• Component catalog Concept · Generative UI & voice

  The set of allowed UI components and their property contracts.

• Computer use Concept · Coding-agent internals

  Operate an application through a computer interface rather than a purpose-built domain API.

• Confused deputy Concept · Security & reliability

  A privileged component is manipulated into acting for a party that lacks the relevant authorization.

• Constitution Concept · Goals, specs & plans

  In Spec Kit, persistent project principles that guide downstream development artifacts.

• Constitutional AI Concept · Models, reasoning & training

  An alignment approach using explicit principles to guide critique and AI-derived feedback.

• Constrained decoding Concept · Tools & output contracts

  Restrict token generation to choices allowed by a grammar or other structural rules.

• Constraint Concept · Goals, specs & plans

  A restriction on permissible behavior or solutions.

• Context budget Concept · Context & memory

  The allocation of limited context capacity across instructions, history, tools, evidence, and output.

• Context isolation Concept · Multi-agent coordination

  Keep task-specific history and working information out of unrelated agents' contexts.

• Context poisoning Concept · Context & memory

  Introduce misleading or malicious information into the material a model uses.

• Context pruning Concept · Context & memory

  Remove information that is no longer useful for the current task.

• Context reset Concept · Context & memory

  Start a fresh context while explicitly carrying forward selected persistent state.

• Context rot Concept · Context & memory

  An informal label for reduced effectiveness as context becomes long, noisy, or distracting.

• Context window Concept · Context & memory

  The amount of context a model can accept within its supported input/output limits.

• Continuous batching Concept · Inference & performance

  Dynamically batch requests as they enter and leave inference execution.

• Control plane Concept · Agents & architecture

  The part of a system that configures, schedules, authorizes, and coordinates execution.

• Copilot Concept · Agents & architecture

  An assistant-oriented interaction pattern in which a human remains actively involved in the work.

• Correctness Concept · Evals & observability

  Whether an answer or action satisfies the actual problem's requirements or truth conditions.

• Correlated failures Concept · Multi-agent coordination

  Multiple agents fail for shared reasons, such as the same missing evidence or model biases.

• Cost optimization Concept · Evals & observability

  Reducing execution spend while preserving the required outcome quality, latency, reliability, and coverage.

• Counterexample Concept · Loops, critics & adversaries

  A concrete case that disproves a general claim or violates a proposed invariant.

• CPU/GPU offloading Concept · Inference & performance

  Move model state or computation between CPU and GPU resources.

• CrewAI Tool · Tools & ecosystem

  A framework for role-based agent collaboration and structured flows.

• Cross-encoder Concept · Retrieval & knowledge

  A model that jointly processes a query and candidate to estimate their relationship.

• Cursor Tool · Tools & ecosystem

  An AI-oriented code editor with assistance and agent workflows.

• Data exfiltration Concept · Security & reliability

  Unauthorized movement or disclosure of information through outputs, tools, or network channels.

• Data lineage Concept · Retrieval & knowledge

  The chain of transformations and dependencies through which data reached its current state.

• Data plane Concept · Agents & architecture

  The part of a system that carries out requests and processes operational data.

• Decode Concept · Inference & performance

  Generate subsequent output tokens using the model and available state.

• Deep agent Concept · Agents & architecture

  An implementation-dependent label often associated with planning, filesystem access, subagents, and context management.

• Deep Agents Tool · Tools & ecosystem

  A LangChain agent harness combining planning, filesystem access, subagents, and context-management facilities.

• Deferred tool loading Concept · Tools & output contracts

  Load full tool definitions only after they become relevant.

• Definition of done Concept · Goals, specs & plans

  A team's completion contract, including checks such as tests, documentation, review, and release readiness.

• Deletion propagation Concept · Retrieval & knowledge

  Ensure removed source content also disappears from indexes, caches, and derived stores as required.

• Dense model Concept · Models, reasoning & training

  A model whose main parameters are generally used for each token, unlike sparse expert activation.

• Dense retrieval Concept · Retrieval & knowledge

  Retrieve candidates using dense vector representations.

• Design by contract Concept · Goals, specs & plans

  Specify preconditions, postconditions, and invariants at software boundaries.

• Deterministic grader Concept · Evals & observability

  A code-based check with explicitly defined decision logic.

• Deterministic workflow Concept · Agents & architecture

  Application code determines the permitted execution paths, even when individual steps use models.

• Developer instructions Concept · Context & memory

  Application-level instructions provided through a platform's developer role or equivalent mechanism.

• Diarization Concept · Generative UI & voice

  Assign segments of audio to distinct speaker identities or labels.

• Diffusers Tool · Tools & ecosystem

  Hugging Face tooling for diffusion and related generative-model pipelines.

• Diffusion model Concept · Generative & physical AI

  A generative model trained to reverse a progressive noising process.

• Distillation Concept · Models, reasoning & training

  Train a student model using supervision obtained from a teacher model or system.

• DOM snapshot Concept · Coding-agent internals

  A representation of webpage structure or accessibility information used for inspection and automation.

• DPO Concept · Models, reasoning & training

  Direct Preference Optimization: train on preference pairs without a conventional separate online RL optimization stage.

• DSPy Tool · Tools & ecosystem

  A framework for programming and optimizing language-model pipelines against chosen metrics.

• Dynamic toolset Concept · Tools & output contracts

  An available set of operations that changes with task, role, permissions, or execution state.

• EARS Concept · Goals, specs & plans

  Easy Approach to Requirements Syntax: structured sentence patterns for stating requirements.

• Embodied AI Concept · Generative & physical AI

  AI that perceives and acts through an agent situated in an environment.

• End-to-end latency Concept · Inference & performance

  Total request or task duration, including queues, model calls, tools, retrieval, and retries.

• Endpointing Concept · Generative UI & voice

  Decide when an incoming utterance has ended or should be finalized.

• Ensemble Concept · Multi-agent coordination

  Combine predictions or outputs from multiple runs or models; it need not involve agent collaboration.

• Episodic memory Concept · Context & memory

  Records of previous events, attempts, observations, and outcomes.

• Eval suite Concept · Evals & observability

  A set of cases covering relevant capabilities, risks, and failure modes.

• Eval-driven development Concept · Evals & observability

  Use systematic evaluation results to guide application and model-integration changes.

• Evaluation frameworks Concept · Evals & observability

  Software or process that organizes evaluation cases, runners, graders, metrics, and reporting for model or application behavior.

• Event-driven agent Pattern · Loops, critics & adversaries

  An agent that starts or resumes in response to an external event.

• Excessive agency Concept · Security & reliability

  Give an agent more functionality, permissions, or autonomy than the task requires.

• Executable specification Concept · Goals, specs & plans

  Requirements expressed through runnable examples or machine-checkable properties.

• Executor Concept · Multi-agent coordination

  A role that carries out assigned actions or implementation work.

• Expert parallelism Concept · Inference & performance

  Distribute mixture-of-experts components across devices.

• Faithfulness Concept · Evals & observability

  Whether an answer remains supported by the evidence supplied to it.

• Fan-out / fan-in Pattern · Multi-agent coordination

  Run independent branches concurrently, then combine their results.

• Fast path / slow path Pattern · Inference & performance

  Separate low-latency handling from more expensive reasoning or processing.

• Few-shot prompting Pattern · Context & memory

  Provide a small set of examples that demonstrate the desired behavior.

• Fine-tuning Concept · Models, reasoning & training

  Adapting pretrained model weights with task, domain, or preference data; supervised fine-tuning and parameter-efficient methods are distinct approaches.

• FlashAttention Concept · Inference & performance

  An IO-aware exact attention algorithm designed to reduce memory traffic and improve execution efficiency.

• Flow matching Concept · Generative & physical AI

  Train a vector field describing a transformation between probability distributions.

• Functional requirements Concept · Goals, specs & plans

  Requirements describing what a system must do.

• Generator-evaluator loop Pattern · Loops, critics & adversaries

  Generate a candidate, assess it against criteria, and feed findings into another attempt.

• Git worktree Concept · Coding-agent internals

  A separate working directory linked to a repository; it is not a security isolation boundary.

• GitHub Copilot Tool · Tools & ecosystem

  GitHub's AI coding assistance and agent-oriented development tooling.

• GitHub Spec Kit Tool · Tools & ecosystem

  A toolkit for specification-driven development artifacts and workflows.

• Glob Concept · Coding-agent internals

  A pattern-based file discovery operation.

• Goal Concept · Goals, specs & plans

  The outcome the system or agent is trying to achieve.

• Goal hijacking Concept · Security & reliability

  Redirect an agent from its authorized objective toward another objective.

• Goal-directed execution Concept · Loops, critics & adversaries

  Select and revise actions by reference to an objective rather than a fixed response template.

• Golden dataset Concept · Evals & observability

  A curated reference set with expected outcomes, properties, or grading criteria.

• Google ADK Tool · Tools & ecosystem

  Google's Agent Development Kit for building agent applications and workflows.

• Graph engineering Concept · Retrieval & knowledge

  An informal umbrella for designing graph-shaped representations, dependencies, or knowledge structures used by AI systems.

• GraphRAG Concept · Retrieval & knowledge

  Use graph-derived structure to support retrieval and synthesis; implementations vary.

• Grep Concept · Coding-agent internals

  A text-search operation over file contents.

• GRPO Concept · Models, reasoning & training

  Group Relative Policy Optimization: estimate relative advantages from groups of sampled responses for policy optimization.

• GSD Tool · Tools & ecosystem

  Get Shit Done: a context-management and structured-development system for coding agents.

• Guardrail Concept · Security & reliability

  A check or constraint on input, output, or actions; its strength depends on where and how it is enforced.

• Hallucination Concept · Evals & observability

  Generated content that is unsupported or incorrect despite being presented as an answer.

• Handoff artifact Concept · Coding-agent internals

  A durable record of progress, decisions, unresolved issues, and next actions.

• Heartbeat Concept · Loops, critics & adversaries

  A periodic trigger that allows a system to inspect state and decide whether work is needed.

• Hermes Agent Tool · Tools & ecosystem

  Nous Research's agent system incorporating tools, persistent memory, and reusable skills.

• Hierarchical agents Pattern · Multi-agent coordination

  Organize agents in multiple levels of planning, supervision, and execution.

• Hill-climbing loop Pattern · Loops, critics & adversaries

  Propose a change, evaluate it, and retain it when it improves the chosen objective.

• HNSW Concept · Retrieval & knowledge

  Hierarchical Navigable Small World: a graph-based approximate nearest-neighbor index.

• Hook Concept · Coding-agent internals

  Code triggered at a lifecycle event, such as before or after a tool call.

• Hugging Face Transformers Tool · Tools & ecosystem

  A library for model implementations, loading, inference, and training.

• Human-in-the-loop Pattern · Multi-agent coordination

  A human participates at a specified decision, review, or approval stage.

• Human-on-the-loop Pattern · Multi-agent coordination

  A human supervises ongoing execution and can intervene without approving every step.

• HyDE Concept · Retrieval & knowledge

  Hypothetical Document Embeddings: generate a hypothetical answer document and use its embedding for retrieval.

• In-context learning Concept · Context & memory

  Adapt behavior from supplied context without updating model weights.

• Index freshness Concept · Retrieval & knowledge

  How well indexed information reflects the current source state.

• Indirect prompt injection Concept · Security & reliability

  A redirection attempt carried in external content, such as retrieved pages, documents, or tool results.

• Instruction hierarchy Concept · Context & memory

  Rules that determine which instructions take precedence when messages conflict.

• Instruction-tuned model Concept · Models, reasoning & training

  A model trained on examples of responding to instructions.

• Invariant Concept · Goals, specs & plans

  A property required to remain true across relevant states and operations.

• ITL Concept · Inference & performance

  Inter-token latency: elapsed time between consecutive emitted tokens.

• Jailbreak Concept · Security & reliability

  An attempt to bypass a model's behavioral restrictions.

• JEPA Concept · Generative & physical AI

  Joint Embedding Predictive Architecture: predict target representations within a learned embedding space.

• JSON mode Concept · Tools & output contracts

  An API mode targeting valid JSON, without necessarily enforcing your complete business schema.

• JSON Schema Concept · Tools & output contracts

  A vocabulary for specifying the structure and constraints of JSON data.

• json-render Tool · Tools & ecosystem

  A framework for rendering model-produced structured UI through constrained component catalogs.

• JSON-RPC Protocol · Protocols & interoperability

  A remote-procedure-call message format encoded with JSON.

• Judge calibration Concept · Evals & observability

  Compare a model judge with trusted assessments and investigate disagreements or biases.

• Just-in-time context Concept · Context & memory

  Load detailed information when needed rather than placing everything in the initial prompt.

• Kiro Tool · Tools & ecosystem

  Development tooling with explicit requirements, design, and task-oriented workflows.

• LangChain Tool · Tools & ecosystem

  A library ecosystem for model integrations, tools, and agent application abstractions.

• Langfuse Tool · Tools & ecosystem

  Open-source observability and evaluation tooling for language-model applications.

• LangGraph Tool · Tools & ecosystem

  Stateful graph-based orchestration with persistence and controlled workflow execution.

• LangSmith Tool · Tools & ecosystem

  Tooling for tracing, evaluation, and agent application development and operations.

• Latent diffusion Concept · Generative & physical AI

  Perform diffusion-based generation in a compressed representation rather than directly in the original data space.

• llama.cpp Tool · Tools & ecosystem

  A C/C++ inference project supporting efficient local execution of compatible model formats.

• LlamaIndex Tool · Tools & ecosystem

  Data ingestion, indexing, retrieval, and knowledge-oriented AI application tooling.

• LLMOps Concept · Agents & architecture

  Operational practices for deploying, evaluating, monitoring, and changing language-model applications.

• Logits Concept · Models, reasoning & training

  Unnormalized model scores over candidate outputs, often transformed into probabilities.

• Long-horizon agent Concept · Agents & architecture

  An agent performing many dependent steps, potentially across sessions or context resets.

• Long-term memory Concept · Context & memory

  Information retained and retrieved across tasks or conversations.

• Loop engineering Concept · Loops, critics & adversaries

  Design the execution, verification, triggering, and improvement loops surrounding an agent.

• LSP Protocol · Coding-agent internals

  Language Server Protocol: standardized communication between editors and language-analysis services.

• MCP client Concept · Protocols & interoperability

  The protocol component communicating with an MCP server.

• MCP elicitation Concept · Protocols & interoperability

  A protocol mechanism for a server to request user-provided information through the client.

• MCP host Concept · Protocols & interoperability

  An application that coordinates its AI interaction and MCP client connections.

• MCP plugin Concept · Coding-agent internals

  A product-dependent integration exposing or connecting MCP capabilities; not a universal packaging format.

• MCP prompts Concept · Protocols & interoperability

  Reusable prompt templates exposed by an MCP server.

• MCP resources Concept · Protocols & interoperability

  Content exposed through MCP for an application to access as context.

• MCP sampling Concept · Protocols & interoperability

  A protocol mechanism for requesting model generation through a client.

• MCP server Concept · Protocols & interoperability

  A service or process exposing capabilities through MCP.

• MCP tools Concept · Protocols & interoperability

  Operations exposed by an MCP server for a client to invoke.

• Memory consolidation Concept · Context & memory

  Transform accumulated experiences into selected, reusable persistent information.

• Memory layers Concept · Context & memory

  An informal architecture shorthand for separating memory by retention, scope, or function, such as working, episodic, semantic, and procedural memory.

• Memory poisoning Concept · Security & reliability

  Corrupt persistent information so it influences later model behavior.

• Memory scoping Concept · Context & memory

  Partition memory by user, tenant, project, task, or permission boundary.

• Metadata filtering Concept · Retrieval & knowledge

  Restrict candidates by structured attributes such as date, tenant, source, or document type.

• Microsoft Agent Framework Tool · Tools & ecosystem

  Microsoft's framework for agent applications and orchestrated workflows.

• Mixture of experts Concept · Models, reasoning & training

  A model architecture routing computation through a subset of expert components.

• Model cascade Pattern · Inference & performance

  Try a less expensive path first and escalate when specified criteria require it.

• Multi-agent debate Pattern · Multi-agent coordination

  Have multiple agents challenge or compare candidate answers before selecting or synthesizing an outcome.

• Multi-agent systems Concept · Multi-agent coordination

  A system in which multiple model-driven components coordinate to complete a shared or decomposed objective.

• Multimodal model Concept · Generative UI & voice

  A model that processes or generates more than one data modality.

• Nonfunctional requirements Concept · Goals, specs & plans

  Requirements for qualities such as latency, reliability, accessibility, and security.

• Observability Concept · Evals & observability

  The ability to infer system behavior from instrumentation such as logs, metrics, traces, and artifacts.

• Offline eval Concept · Evals & observability

  Evaluate controlled examples outside live user traffic.

• Ollama Tool · Tools & ecosystem

  Tooling for running and managing supported language models through local interfaces and APIs.

• Online eval Concept · Evals & observability

  Assess behavior observed during production use.

• Open weights Concept · Models, reasoning & training

  Model weights are accessible under a license; training data and unrestricted usage rights are not implied.

• OpenAI Agents SDK Tool · Tools & ecosystem

  A library for agents, tools, handoffs, and execution tracing.

• OpenClaw Tool · Tools & ecosystem

  A personal-assistant system integrating tools, sessions, channels, and automation.

• OpenSpec Tool · Tools & ecosystem

  A specification-centered workflow for proposing and implementing software changes with AI assistants.

• OpenTelemetry Concept · Evals & observability

  A vendor-neutral framework and specifications for producing and exporting observability data.

• Orchestration Concept · Agents & architecture

  Coordinating tasks, agents, tools, dependencies, and shared execution state.

• Orchestrator Concept · Multi-agent coordination

  A coordinator that assigns work, manages dependencies, and synthesizes results.

• Paged attention Concept · Inference & performance

  Manage attention KV-cache memory in blocks to improve allocation and sharing efficiency.

• Parallel tool calls Concept · Tools & output contracts

  Execute independent operations concurrently, subject to permissions and dependency constraints.

• Parent-child retrieval Concept · Retrieval & knowledge

  Match a smaller passage, then retrieve its larger containing context.

• pass@k Concept · Evals & observability

  The probability or estimated rate that at least one of k sampled attempts succeeds.

• pass^k Concept · Evals & observability

  A consistency measure asking whether all k trials succeed under the evaluation's definition.

• Patch / edit tool Concept · Coding-agent internals

  A controlled file-change operation that modifies selected content rather than regenerating the entire artifact.

• PEFT Concept · Models, reasoning & training

  Parameter-efficient fine-tuning: adapt a model by training a limited subset or added parameters.

• PEFT library Tool · Tools & ecosystem

  Hugging Face tooling for parameter-efficient model adaptation.

• Permission-aware retrieval Concept · Retrieval & knowledge

  Enforce access restrictions when selecting and returning source material.

• Persistent task execution Concept · Loops, critics & adversaries

  Continue a task across interruptions using stored progress and an explicit completion contract.

• pgvector Tool · Tools & ecosystem

  A PostgreSQL extension for storing vectors and performing similarity searches.

• Physical AI Concept · Generative & physical AI

  AI interacting with physical systems, often through sensors, actuators, simulation, and learned policies.

• Pi Tool · Tools & ecosystem

  An agent toolkit that includes model interfaces, agent execution components, and a coding-agent interface.

• Plan mode Concept · Goals, specs & plans

  A product-specific mode intended for exploration and planning before implementation; restrictions vary by tool.

• Plan-and-execute Pattern · Loops, critics & adversaries

  Construct an explicit plan, then execute its steps and track progress.

• Planner Concept · Multi-agent coordination

  A role responsible for structuring the approach and decomposing tasks.

• Playwright Tool · Tools & ecosystem

  A browser automation and testing library used to verify real application behavior.

• Post-training Concept · Models, reasoning & training

  Training after pretraining to improve instruction following, preferences, specialized behavior, or reasoning.

• PPO Concept · Models, reasoning & training

  Proximal Policy Optimization: a policy-gradient reinforcement-learning method using constrained updates.

• PRD Concept · Goals, specs & plans

  A product requirements document describing users, problems, intended behavior, scope, and constraints.

• Precision / recall Concept · Evals & observability

  Measures of how many selected items are relevant and how many relevant items were found.

• Prefill Concept · Inference & performance

  Process input tokens to establish the state used for subsequent generation.

• Prefill-decode disaggregation Concept · Inference & performance

  Manage input processing and token generation on separately allocated serving resources.

• Prefix caching Concept · Inference & performance

  Reuse computation for matching input prefixes.

• Pretraining Concept · Models, reasoning & training

  Broad training that establishes a model's initial representations and capabilities.

• Procedural memory Concept · Context & memory

  Stored instructions describing how to perform tasks.

• Prompt chaining Pattern · Context & memory

  Feed the output of one model step into a subsequent model step.

• Prompt engineering Concept · Context & memory

  Design instructions, examples, and output requests to guide model behavior.

• Prompt optimization Pattern · Context & memory

  Systematically improving prompts against a chosen objective using examples, evaluation, search, or programmatic optimization.

• Promptfoo Tool · Tools & ecosystem

  Tooling for prompt and application evaluations and security-oriented testing.

• Provenance Concept · Retrieval & knowledge

  Information about where a fact or artifact originated.

• Pydantic Tool · Tools & output contracts

  A Python library for data validation and serialization using type annotations.

• Qdrant Tool · Tools & ecosystem

  A vector search system with filtering and hybrid-retrieval capabilities.

• QLoRA Concept · Models, reasoning & training

  Parameter-efficient adaptation using low-rank updates with a quantized base model.

• Quantization Concept · Inference & performance

  Represent model values with reduced precision to lower memory or compute requirements.

• Query decomposition Concept · Retrieval & knowledge

  Split a complex information need into multiple retrievable subquestions.

• Query rewriting Concept · Retrieval & knowledge

  Reformulate a request to make retrieval more effective.

• RAG 2.0 Pattern · Retrieval & knowledge

  An informal label for an iteration of retrieval-augmented generation that claims improved retrieval, indexing, agent behavior, or evaluation.

• Ralph loop Pattern · Loops, critics & adversaries

  An informal coding-agent pattern of repeated invocations with progress carried through persistent repository artifacts.

• Rate limit Concept · Inference & performance

  A restriction on request, token, concurrency, or resource consumption over a defined interval.

• ReAct Pattern · Loops, critics & adversaries

  A prompting pattern interleaving reasoning, actions, and observations.

• Read Concept · Coding-agent internals

  A tool primitive that returns selected file content.

• Realtime agent Concept · Generative UI & voice

  An agent designed for ongoing low-latency interaction rather than only completed batch responses.

• Reasoning budget Concept · Inference & performance

  A configured limit or allocation for inference-time reasoning effort.

• Red teaming Concept · Security & reliability

  Systematically search for security, safety, and misuse failures within an authorized scope.

• Reflection Concept · Loops, critics & adversaries

  A model evaluates its own previous output or behavior to inform what to do next.

• Reflexion Pattern · Loops, critics & adversaries

  A research method using verbal feedback and episodic memory to improve subsequent attempts.

• Regression eval Concept · Evals & observability

  Check whether a change harms previously acceptable behavior.

• Replay Concept · Security & reliability

  Reconstruct or resume execution from previously recorded state or events.

• Repository map Concept · Coding-agent internals

  A compact structural overview of important files, symbols, or dependencies.

• Requirements traceability Concept · Goals, specs & plans

  Links from requirements to implementation tasks, code, and verification evidence.

• Retrieval policy Concept · Context & memory

  Rules for deciding what external information to fetch and what to include in context.

• Retry loop Pattern · Loops, critics & adversaries

  Repeat an operation after failure, often without changing the underlying candidate or strategy.

• Retry policy Concept · Security & reliability

  Rules for whether, when, and how often failed operations should be attempted again.

• Reward hacking Concept · Security & reliability

  Optimize a reward signal in a way that defeats its intended purpose.

• Reward model Concept · Models, reasoning & training

  A model that estimates preference or quality for training, evaluation, or candidate selection.

• RLAIF Concept · Models, reasoning & training

  Reinforcement learning using feedback generated by AI systems.

• RLHF Concept · Models, reasoning & training

  Reinforcement learning using reward signals derived from human feedback.

• RLVR Concept · Models, reasoning & training

  Reinforcement learning with verifiable rewards, such as outcomes checked by tests or objective graders.

• Router Concept · Multi-agent coordination

  Select an agent, model, tool, or workflow based on the request and current state.

• RRF Concept · Retrieval & knowledge

  Reciprocal rank fusion: combine ranked result lists using rank-based scores.

• Rubric Concept · Evals & observability

  Explicit criteria defining how an output is evaluated.

• Rules / steering files Concept · Coding-agent internals

  Product-specific instructions scoped to a project, file type, situation, or workflow.

• Sandbox Concept · Coding-agent internals

  An execution environment with enforced isolation and access or resource restrictions.

• Sandbox escape Concept · Security & reliability

  Circumvent an execution environment's intended isolation boundary.

• Scaffold Concept · Agents & architecture

  Supporting code and structure around model execution; its boundary overlaps with harness in some usage.

• Schema validation Concept · Tools & output contracts

  Check that data satisfies its structural contract.

• Schema-driven UI Concept · Generative UI & voice

  Render an interface from structured data constrained by a defined schema.

• Scratchpad Concept · Context & memory

  Temporary working information used during a task; it may exist as explicit files or application state.

• Self-consistency Pattern · Models, reasoning & training

  Aggregate results from multiple sampled reasoning attempts, often by answer agreement.

• Self-improving agent Concept · Loops, critics & adversaries

  An ambiguous label: improvement may change prompts, memory, skills, or weights; specify which.

• Self-refinement Pattern · Loops, critics & adversaries

  Repeatedly revise an output using feedback generated internally or externally.

• Semantic cache Concept · Inference & performance

  Reuse a previous result for a sufficiently similar request, subject to validity and permission checks.

• Semantic Kernel Tool · Tools & ecosystem

  A Microsoft SDK for integrating models, tools, and AI-oriented orchestration.

• Semantic memory Concept · Context & memory

  Stored factual knowledge or relationships.

• Semantic search Concept · Retrieval & knowledge

  Search by represented meaning rather than only literal word overlap.

• Semantic validation Concept · Tools & output contracts

  Check that values are meaningful and satisfy domain rules, beyond structural validity.

• SGLang Tool · Tools & ecosystem

  A model-serving system with optimized inference and scheduling capabilities.

• Shared state Concept · Multi-agent coordination

  Data that multiple participants can read or update during coordination.

• Shell / terminal tool Concept · Coding-agent internals

  An operation for running commands, builds, tests, and local programs.

• SKILL.md File convention · Coding-agent internals

  A skill's metadata and procedural instructions, optionally accompanied by scripts and reference files.

• Slash command Concept · Coding-agent internals

  A user-facing shortcut that invokes a prompt, skill, or workflow.

• SLM Concept · Models, reasoning & training

  Small language model: a relative size label with no single universal parameter threshold.

• Source attribution Concept · Retrieval & knowledge

  Associate claims with the source material supporting them.

• Span Concept · Evals & observability

  A timed operation within a distributed or application trace.

• Sparse retrieval Concept · Retrieval & knowledge

  Retrieve using sparse term or feature representations.

• Spec drift Concept · Goals, specs & plans

  A mismatch between the stated specification and the implemented behavior.

• Specification gaming Concept · Security & reliability

  Satisfy literal measured criteria while violating the intended objective.

• Speculative decoding Concept · Inference & performance

  Use a cheaper draft mechanism to propose tokens that a target model verifies.

• Speculative execution Concept · Inference & performance

  Start potentially useful application work before knowing whether it will be needed.

• Speech-to-speech Concept · Generative UI & voice

  Generate spoken output from spoken input, with architecture-specific intermediate processing.

• SSE Protocol · Protocols & interoperability

  Server-Sent Events: an HTTP mechanism for streaming server-to-client events.

• Stagnation detection Concept · Loops, critics & adversaries

  Identify repeated work or errors without meaningful progress.

• Stateless MCP Concept · Protocols & interoperability

  An informal shorthand for serving MCP requests without retaining transport session state between requests.

• stdio Concept · Protocols & interoperability

  A local process transport using standard input and standard output.

• Strands Agents Tool · Tools & ecosystem

  An AWS-originated SDK for building model-driven agents with tools.

• Streamable HTTP Concept · Protocols & interoperability

  An HTTP-based MCP transport supporting protocol communication and optional streaming behavior.

• Streaming state Concept · Generative UI & voice

  Deliver structured updates incrementally while execution is still in progress.

• Success criteria Concept · Goals, specs & plans

  Observable conditions that establish whether the intended outcome was achieved.

• Superpowers Tool · Tools & ecosystem

  A skills-based collection of software-development workflows for coding agents.

• Supervisor Concept · Multi-agent coordination

  An oversight role that routes work, checks progress, and may intervene or escalate.

• Swarm Concept · Multi-agent coordination

  An informal label for a collection of cooperating agents; the coordination rules still need specification.

• Symbol indexing Concept · Coding-agent internals

  Index names, definitions, references, and relationships in a codebase.

• Synthetic data Concept · Models, reasoning & training

  Generated examples used for training, evaluation, or augmentation.

• System 1 / System 2 Concept · Models, reasoning & training

  Informal metaphors for fast direct responses and deliberative processing, not standard model architectures.

• System instructions Concept · Context & memory

  High-priority instructions governing the assistant's behavior; exact role semantics depend on the platform.

• Tail latency Concept · Inference & performance

  The slow end of a latency distribution, commonly summarized by p95 or p99.

• Task DAG Concept · Goals, specs & plans

  A directed acyclic graph of task dependencies; cycles require separate iterative control logic.

• Task decomposition Concept · Goals, specs & plans

  Breaking an objective into smaller work units with clear boundaries.

• Technical design Concept · Goals, specs & plans

  The architecture, interfaces, data structures, and trade-offs chosen to satisfy requirements.

• Temperature Concept · Models, reasoning & training

  A sampling parameter that reshapes relative token probabilities; its effect depends on the decoding setup.

• Tensor parallelism Concept · Inference & performance

  Split operations within model layers across multiple devices.

• Timeout Concept · Security & reliability

  A limit after which an operation is considered expired or interrupted according to its execution contract.

• Tokenization Concept · Models, reasoning & training

  Convert input into units from a model-specific vocabulary.

• Tool allowlist / denylist Concept · Coding-agent internals

  Explicitly permitted or prohibited operations enforced by a runtime or policy layer.

• Tool grounding Concept · Tools & output contracts

  Base a requested operation on real, available capabilities and observed identifiers.

• Tool poisoning Concept · Security & reliability

  Malicious or misleading instructions placed in tool descriptions, metadata, or related tool-facing content.

• Tool schema Concept · Tools & output contracts

  A machine-readable contract describing an operation's inputs and, sometimes, outputs.

• Tool selection Concept · Tools & output contracts

  Choose which available operation is appropriate for the current task.

• Tool use Concept · Tools & output contracts

  Using external operations, services, or interfaces as part of completing a task.

• Tool-result shaping Concept · Tools & output contracts

  Return compact, relevant, well-structured observations rather than unfiltered raw data.

• Top-p Concept · Models, reasoning & training

  Nucleus sampling: select from a probability-ranked token set meeting a cumulative probability threshold.

• Total parameters Concept · Models, reasoning & training

  The full parameter count of a model, including components not active on every token.

• TPOT Concept · Inference & performance

  Time per output token: an average token-generation latency under a specified measurement method.

• TPS Concept · Inference & performance

  Tokens per second; specify whether it measures one request or aggregate system throughput.

• Trajectory Concept · Evals & observability

  The sequence of actions and observations in an execution run.

• Trajectory evaluation Concept · Evals & observability

  Assess the execution path, not only the final answer.

• Transformer Concept · Models, reasoning & training

  A neural architecture built around attention and learned transformations of token representations.

• TRL Tool · Tools & ecosystem

  Hugging Face tooling for supervised and preference or reinforcement-learning post-training.

• TTS Concept · Generative UI & voice

  Text-to-speech: generate spoken audio from text.

• Turn detection Concept · Generative UI & voice

  Determine conversational turn boundaries so an agent responds at an appropriate time.

• UI action binding Concept · Generative UI & voice

  Connect a rendered control to a permitted application operation.

• VAD Concept · Generative UI & voice

  Voice activity detection: identify intervals that contain speech.

• Vector database Concept · Retrieval & knowledge

  A data system designed to store vectors and support similarity retrieval, often with metadata filters.

• Vercel AI SDK Tool · Tools & ecosystem

  TypeScript tooling for model integration, streaming, tools, and AI application interfaces.

• Verification loop Pattern · Loops, critics & adversaries

  Repeat implementation and checks until acceptance or an execution limit is reached.

• Vibe coding Concept · Agents & architecture

  An informal style of building through conversational AI direction, often without closely inspecting generated implementation.

• VLA Concept · Generative & physical AI

  Vision-language-action model: map visual observations and language to actions.

• vLLM Tool · Tools & ecosystem

  A model inference and serving engine focused on efficient execution and throughput.

• VLM Concept · Generative UI & voice

  Vision-language model: a model combining visual and language capabilities.

• Watchdog Concept · Loops, critics & adversaries

  An independent monitor that detects and interrupts stalled or unsafe execution.

• WebSocket Protocol · Protocols & interoperability

  A persistent, bidirectional communication channel commonly used for realtime application traffic.

• Windsurf Tool · Tools & ecosystem

  An AI-oriented development environment with code assistance and agent workflows.

• Working memory Concept · Context & memory

  Task-local information currently available or selected for the ongoing interaction.

• World model Concept · Generative & physical AI

  A learned representation or predictive model of an environment and its dynamics.

• Zero-shot prompting Pattern · Context & memory

  Request a task without supplying task-specific demonstration examples.

• Zod Tool · Tools & output contracts

  A TypeScript-oriented runtime schema and validation library.

Review cards · Distinctions · Cheat sheets