A
Abuse monitoring
Azure OpenAI process that detects patterns of misuse across a customer's traffic for Microsoft review; it does not filter individual responses.
Accountability principle
Responsible AI principle that named people remain answerable for AI, with human oversight, the ability to override and governance committees.
Adaptive thinking (thinking type adaptive)
Claude thinking mode where the model decides whether and how much to reason, steered by the effort parameter; the only thinking mode on Claude Opus 4.7 and later models.
ADLS Gen2 (Azure Data Lake Storage Gen2)
Standard GPv2 storage with hierarchical namespace enabled, giving real directories and POSIX-style ACLs for analytics.
Adult content detection (Adult visual feature)
Image Analysis 3.2 feature that returns isAdultContent, isRacyContent and isGoryContent flags with 0–1 scores, with no training; Content Safety is the newer moderation offering.
Advanced data parsing
Parsing that runs OCR on scans, merges multipage tables, restores headers and creates structure-aware chunks with heading, page and table metadata, unlike fixed-size or per-page chunking.
Affordance (prompting)
Prompt technique of giving the model tools or functions it can call, so it hands work such as lookups or calculations to code instead of guessing.
Agent Application (published agent)
Azure resource created when you publish an agent version, with a stable endpoint and its own Entra agent identity and blueprint; RBAC granted to the project identity doesn't transfer and must be reassigned.
Agent identity
Special service principal for an AI agent with no credentials of its own; it acts autonomously with app-only permissions or on behalf of a user with delegated permissions, and Conditional Access can only block it (no grant controls).
Agent identity blueprint (blueprint)
Microsoft Entra object that is the template and credential holder for a kind of agent identity; policies such as Conditional Access applied to it cover every agent identity created from it.
Agent instructions (system message)
The agent's system message: sets its role, goals, tone and limits and guides (non-deterministically) when it uses tools; deployment slots, embeddings or fine-tuning don't define the role.
Agent Monitoring Dashboard
Foundry monitoring view backed by Application Insights showing token usage, latency, run success rate and evaluation scores, with alerts; runtime visibility, not a release gate.
Agent version
Immutable snapshot of an agent's configuration; any change, even one prompt edit, creates a new version, and rollback means pointing at a prior version.
Agentic retrieval
Azure AI Search pipeline where an LLM plans subqueries from the conversation, runs them in parallel with semantic reranking and merges the best chunks; classic RAG sends one query.
Agents (classic) API (threads, runs, messages)
Original Foundry Agent Service API built on threads, runs and messages; deprecated and retiring 31 March 2027 in favour of conversations and responses.
AI agent
Application that uses a generative model with tools to interpret requests, hold a dialogue and take actions toward a goal.
AI enrichment (skillset)
Azure AI Search pipeline in which skills (OCR, entity recognition, key phrases, image analysis) transform raw content into searchable fields during indexing.
AI gateway (Foundry)
Azure API Management instance associated with a Foundry resource that proxies registered agents, tools and models and adds access control, rate limits and diagnostics; a prerequisite for registering custom agents.
AIProjectClient
Foundry SDK entry point created with the project endpoint and an Entra credential such as DefaultAzureCredential; Entra ID is the only supported auth, so there is no project key or SAS.
AIServices (resource kind)
Kind value for the Microsoft Foundry multi-service resource, the recommended kind, giving Azure OpenAI, Speech, Vision, Language and Content Understanding behind one endpoint and credential.
Alt text (Image Analysis)
Accessibility caption for an image, generated by Image Analysis captioning with a confidence threshold (about 0.4 for v3.2); for new work, a multimodal model or Content Understanding is now preferred.
Analyze image API (image:analyze)
Content Safety operation (POST {endpoint}/contentsafety/image:analyze) that scores a base64 image or blob URL for the four harm categories at severities 0, 2, 4 or 6.
Analyze text API (text:analyze)
Content Safety operation (POST {endpoint}/contentsafety/text:analyze) that scores text for hate, sexual, violence and self-harm and can check named blocklists.
analyze-text endpoint (/language/:analyze-text)
Unified Azure Language REST endpoint where the request kind (LanguageDetection, EntityRecognition, SentimentAnalysis and others) selects the feature; it replaces /text/analytics/v3.x paths.
AnalyzeTextOptions
Content Safety SDK request object for text analysis (text, categories, blocklist names, output type); it is passed to AnalyzeText, not AnalyzeImage.
Annotate (guardrail action)
Guardrail action that returns detection results (detected, filtered, severity) without blocking content; use Annotate and block to stop it.
Annotate and block
Foundry guardrail action that flags and blocks detected risk; at the tool call intervention point it stops the tool call from executing.
API key (Foundry resource key) (resource key)
Secret key issued with a Foundry Tools or Azure OpenAI resource and sent in the api-key header; the fallback when keyless Entra ID auth isn't possible.
API Management (APIM)
API gateway applying policies such as validate-jwt, ip-filter, rate limits and quotas once for all APIs; Premium tier for production VNet injection.
APIM subscription key (Ocp-Apim-Subscription-Key)
Key header (Ocp-Apim-Subscription-Key) sent with API calls; for Foundry Tools such as Content Safety it carries the resource key, not the Azure subscription ID, and for API Management it carries the APIM subscription key.
Application Insights
Azure Monitor APM service for app telemetry, Application Map, availability tests and usage analytics; workspace-based instances store data in Log Analytics.
Ask a question node (ask_question)
Foundry workflow node that sends a question, waits for the user's reply and saves it to a variable (for example Local.Var01) for later conditions.
Assistants API (Azure OpenAI Assistants)
Deprecated OpenAI-style assistants, threads and runs API, retired 26 August 2026 and replaced by Foundry agents with conversations and responses.
Audio + human-labeled transcript data
Custom speech training data: a .zip of WAV files (RIFF, 8 or 16 kHz, 16-bit PCM, mono) plus a plain-text file mapping each file name to its transcript.
AudioConfig
Speech SDK object that defines audio input or output, e.g. FromDefaultMicrophoneInput, FromWavFileInput or FromStreamInput.
AudioOutputConfig
Speech SDK class that sets where synthesised audio goes: the default speaker, a file (filename="output.wav"), a custom output stream or a named device, one at a time.
Autocomplete
Azure AI Search query API that completes partial terms from suggester fields, called with suggesterName; Suggestions returns matching documents instead.
AutoGen
Microsoft Research framework for multi-agent systems (such as GroupChat); superseded by Microsoft Agent Framework.
az cognitiveservices account create
Azure CLI command that creates a Foundry or Foundry Tools resource from --kind, --sku and --location (programmatic name such as eastus); customer-managed key settings go in --encryption, and --assign-identity only creates the managed identity.
Azure AI Content Safety
Foundry Tool that detects hate, sexual, violence and self-harm content in text and images with severity levels; the same technology powers Foundry guardrails.
Azure AI Search
Managed search service providing keyword, vector and hybrid search, AI enrichment and retrieval for RAG and knowledge mining.
Azure AI Search admin key
One of two regenerable primary and secondary API keys granting full read-write access to a search service; rotate by switching clients to the secondary, regenerating the primary, then reversing.
Azure AI Search partition (partition)
Unit of storage and I/O holding a slice of the indexes; add partitions for indexing throughput or storage, since replicas × partitions = search units.
Azure AI Search query key
Read-only API key (up to 50 per service) for querying an index's documents collection; query keys can't be regenerated, only created and deleted, and apply to every index.
Azure AI Search replica (replica)
Copy of the search engine and indexes that load-balances queries and provides availability; add replicas for query throttling (HTTP 503).
Azure AI Search tool (AzureAISearchTool)
Agent tool that queries an existing Azure AI Search index, configured with project_connection_id and index_name (default query type vector_semantic_hybrid, top_k 5); the tool must exist before the agent is created.
Azure CLI (CLI)
Cross-platform command line for managing Azure, e.g. az cognitiveservices account create or az login.
Azure Content Moderator (Content Moderator)
Deprecated predecessor of Azure AI Content Safety (the /contentmoderator path) with binary flagging and custom term lists; retires 15 March 2027, so new work uses Content Safety.
Azure Content Understanding in Foundry Tools (Content Understanding)
Foundry Tool that uses generative AI analyzers to extract, classify and generate structured fields from documents, images, audio and video.
Azure Data Explorer
Petabyte-scale interactive analytics for logs and telemetry using KQL; not a supported Azure AI Search indexer data source.
Azure Data Factory
Managed ETL/ELT with pipelines, copy activities, triggers, mapping data flows and integration runtimes; batch, not per-transaction routing.
Azure Document Intelligence in Foundry Tools (Document Intelligence, Form Recognizer)
Foundry Tool with read, layout, prebuilt (invoice, receipt) and custom models that extract text and fields from forms and documents.
Azure Files
Managed SMB/NFS file shares; a knowledge store can't write to Azure Files (only Blob and Table storage).
Azure Firewall
Managed stateful network firewall, deployable in Virtual WAN hubs and managed by Firewall Manager.
Azure Functions
Serverless event-driven code (Consumption, Premium or Dedicated plan) run by triggers such as timer, HTTP, Blob or Event Grid.
Azure Key Vault (Key Vault)
Store for secrets, keys and certificates; where a same-geography pair exists it replicates there, with best-effort Microsoft-initiated failover during which the vault is read-only.
Azure Language in Foundry Tools (Azure AI Language, Text Analytics)
Foundry Tool of NLP features (sentiment, key phrases, NER, PII, summarisation, language detection) that analyse digital text; prebuilt features need no training and it doesn't generate new content.
Azure Login action (azure/login)
GitHub Actions step that signs the workflow in to Azure as an Entra app's service principal or a user-assigned managed identity, ideally via OpenID Connect so no secret is stored.
Azure Machine Learning (AML)
Cloud service for training, deploying and managing your own ML models with MLOps; the wrong answer when a prebuilt language model does the writing.
Azure Machine Learning pipeline (ML pipeline)
Workflow that chains components (for example prepare data, train, score) into a reusable, schedulable job.
Azure Monitor agent (AMA)
Current agent collecting guest-OS logs according to DCRs; replaced the Log Analytics agent (MMA).
Azure OpenAI (Azure OpenAI in Foundry Models)
Azure access to OpenAI models (GPT, embeddings, gpt-image) through a Foundry or Azure OpenAI resource; a standalone Azure OpenAI resource doesn't expose Content Understanding.
Azure OpenAI AvailabilityRate (Model Availability Rate)
Azure OpenAI metric equal to (total calls minus server errors) / total calls, showing service-side unavailability (HTTP 5xx).
Azure OpenAI Embedding skill (AzureOpenAIEmbeddingSkill)
Skill that calls an embedding model deployment (text-embedding-3-small/large or ada-002) to vectorise each chunk during indexing; billed by Azure OpenAI.
Azure OpenAI in Foundry Models (Azure OpenAI Service)
Fully managed Azure access to OpenAI models (GPT-4.1, GPT-5, o-series, image, audio, embeddings) with built-in guardrails; you choose a deployment type instead of managing infrastructure.
Azure OpenAI On Your Data
Deprecated Azure OpenAI feature that grounded chat on your data through an Azure AI Search index; new work uses Foundry Agent Service with Azure AI Search or Foundry IQ.
Azure OpenAI v1 API
Versionless Azure OpenAI API called with the standard OpenAI() client and base_url https://<resource-name>.openai.azure.com/openai/v1/; the host is the resource name, not the project.
Azure Private Link (Private Link)
Platform that exposes PaaS services such as Azure Storage on a private endpoint in your VNet over the Microsoft backbone; needs VNet and DNS setup and does not admit a public IP.
Azure Speech in Foundry Tools (Azure Speech)
Foundry Tool providing speech to text, text to speech, speech translation and live voice conversations.
Azure SQL Database
PaaS single database or elastic pool; up to 4 TB (128 TB Hyperscale); no cross-database queries, SQL Agent or CLR.
Azure SQL Managed Instance (MI)
PaaS SQL Server instance with near-full compatibility (SQL Agent, CLR, cross-database queries); regional DR only via auto-failover groups.
Azure Translator in Foundry Tools (Translator)
Foundry Tool for text and document translation between languages; it doesn't extract fields from forms.
Azure Vision in Foundry Tools (Azure AI Vision, Computer Vision)
Foundry Tool for image analysis, OCR and face detection; the Image Analysis API (v3.2 and v4.0) is deprecated and retires on 25 September 2028.
azure-ai-projects (AIProjectClient)
Microsoft Foundry Python SDK whose AIProjectClient connects to a project; telemetry.get_application_insights_connection_string() returns the project's Application Insights connection string.
azure-core-tracing-opentelemetry
Python package that plugs Azure SDK calls into OpenTelemetry; installed with opentelemetry-sdk and azure-ai-projects for client-side agent tracing.
AzureAIOpenTelemetryTracer
LangChain and LangGraph callback tracer that emits OpenTelemetry spans to Foundry and Azure Monitor only when passed via config={'callbacks':[tracer]}; enable_content_recording controls message content.
AzureKeyCredential
Azure SDK credential that sends a static API key; it breaks a no-keys rule, unlike DefaultAzureCredential.
B
Base analyzer
Modality-specific parent analyzer (prebuilt-document, prebuilt-image, prebuilt-audio, prebuilt-video) that custom analyzers inherit from via baseAnalyzerId.
Batch synthesis API
Asynchronous text to speech REST API for large volumes of plain text or SSML that can produce audio longer than 10 minutes, such as audiobooks; the real-time REST API truncates at 10 minutes.
Batch transcription
Asynchronous speech to text REST API for large volumes of prerecorded audio in storage, with results written back later; not for live captions.
Bearer token (access token)
Access token sent in the HTTP Authorization: Bearer header so that whoever holds it can call the protected API.
Billable Foundry resource (skillset) (cognitiveServices)
Foundry resource key or identity attached to a skillset so built-in skills (Vision, Language, Translator) can exceed the free 20 documents per indexer per day; one multi-service resource covers them all.
BingGroundingToolDefinition
C# (classic agents SDK) tool definition built from the Bing connection ID and passed in the tools list to CreateAgentAsync, not in ToolResources.
Blob storage
Object storage for unstructured data such as video and images; block blobs up to about 190.7 TiB.
Blocklist
Custom list of terms or regex patterns (up to 10,000 per list) attached to a content filter to block specific words; changing it is a content-filter change, not monitoring or evaluation.
BlocklistClient
Content Safety SDK client that creates, lists and deletes text blocklists and adds or removes their items (add_or_update_blocklist_items, remove_blocklist_items up to 100 per call, delete_text_blocklist).
BOM (byte-order mark)
Marker at the start of a text file; fine-tuning files must be UTF-8 with a BOM and each under 512 MB, while ISO-8859-1, ASCII or UTF-16 can fail validation.
Bounding box
Four pixel values (x, y, width, height) that locate a detected object in an image; returned by object detection but not by image classification.
Branch protection
Repository rule (GitHub branch protection or Azure Repos branch policies) that blocks direct pushes to a branch such as main and requires pull request reviews before merging.
Brand detection (Brands)
Image Analysis 3.2-only feature that finds commercial logos from a built-in database and returns brand name, confidence and bounding box; custom logos need a trained model.
Built-in evaluators (evaluators)
Foundry evaluators that measure and report quality, RAG, safety and agent behaviour; they score outputs but don't change settings or retrain the model.
C
Chain-of-thought prompting (chain of thought, CoT)
Prompting technique that asks the model to reason step by step and show each step before the answer; for non-reasoning models only.
Chat Completions API (chat.completions)
Message-based Azure OpenAI API (messages array with roles, response_format, tools with a nested function object); still supported, but new apps should use the Responses API.
Chat message roles (system, user, assistant, tool)
Roles in a chat request: system holds instructions, user the end user's input, assistant earlier model replies, and tool (or function) the output returned from a tool call.
Chat playground
No-code Foundry page for testing a deployed chat model with a system message and parameters such as temperature, max response and past messages.
Chunking
Splitting documents into smaller passages before embedding and indexing; Azure AI Search suggests about 512 tokens with roughly 10-25% overlap, and oversized chunks dilute meaning across unrelated sections.
Claude models in Foundry (Anthropic Claude)
Anthropic models deployable in Microsoft Foundry; newer ones (Opus 4.7 onward) use adaptive thinking only and reject temperature and thinking type enabled, while Sonnet 4.6 still accepts them.
ClientSecretCredential
Azure Identity credential that authenticates as a service principal with a tenant ID, client ID and client secret, so a stored secret must be protected and rotated.
CLU (conversational language understanding)
Azure Language custom feature that predicts the intent of user utterances and extracts entities for chatbots; retiring 31 March 2029.
Code interpreter tool (code_interpreter)
Built-in agent tool that writes and runs Python in a sandbox for calculations, data analysis and charts; not needed to query a search index.
Cognitive Services Contributor
Management role that creates resources, views and regenerates keys and customises content filters, but can't make inference calls with Entra ID; never the least-privilege answer.
Cognitive Services Data Reader
Built-in role with read-only data actions on Foundry Tools data (for example Speech projects).
Cognitive Services OpenAI Contributor
Azure OpenAI role with OpenAI User rights plus uploading datasets, fine-tuning and creating deployments; it can't create resources or view keys.
Cognitive Services OpenAI User
Least-privilege Azure OpenAI role for inference with Microsoft Entra ID, viewing endpoint and deployments, and playgrounds; it can't upload fine-tuning data, deploy or view keys.
Cognitive Services Speech User
Speech role for real-time and batch transcription and synthesis APIs with read-only access to custom models; used with Entra ID auth plus a custom subdomain.
Cognitive Services User
Built-in role that can read and list keys and call all Foundry Tools and models on a resource; broader than needed for Azure OpenAI inference only.
CognitiveServices (resource kind)
Kind value for the older multi-service Azure AI services resource; new multi-service resources should use AIServices.
Completions playground (legacy completions API)
Legacy prompt-in, text-out playground for the old Completions API (for example gpt-35-turbo-instruct); new work uses the Foundry chat playground and the Responses or Chat Completions API.
Composed model
Document Intelligence model grouping up to 500 custom extraction models behind one ID; since v4.0 (2024-11-30) an explicitly trained classifier routes each document.
Compressed audio input (AudioStreamFormat.GetCompressedFormat)
Speech SDK support for MP3, OPUS/OGG, FLAC, ALAW and MULAW input via a push or pull stream, decoded by GStreamer; GetWaveFormatPCM is for uncompressed audio.
Computer use tool (computer use)
Agent tool that lets a model operate a graphical interface (click, type, read screenshots) to automate GUI tasks.
Confidence score
Probability from 0 to 1 attached to each prediction or extracted field, showing how sure the model is; distinct from accuracy, which is measured over a test set.
configure_azure_monitor
Azure Monitor OpenTelemetry distro function that takes an Application Insights connection_string and exports that service's traces, metrics and logs to Azure Monitor.
Consent statement (verbal statement)
Audio recording (.wav or .mp3, not a signed PDF) of the voice talent reading a prescribed statement agreeing to the use of their voice, in the same language and recording conditions as the training data.
Content extraction
First Content Understanding stage that normalises input into text and metadata: OCR and layout for documents, transcription for audio and video.
Content filter (guardrail controls)
Foundry guardrail configuration that classifies prompts and completions by risk (hate, sexual, violence, self-harm, prompt attacks, protected material) and blocks at a severity threshold; not an observability tool.
Content Moderator (Azure Content Moderator)
Legacy moderation service, deprecated in February 2024 and retiring on 15 March 2027; replaced by Azure AI Content Safety.
Content Safety severity levels
Severity scale per harm category: text uses 0–7 (or trimmed 0, 2, 4, 6) and images return only 0, 2, 4 or 6 (safe, low, medium, high).
Content Safety Studio
Online portal for Azure AI Content Safety that tests text and image moderation with adjustable severity thresholds, manages blocklists, exports code and monitors usage; a no-code prototype needs only a Content Safety resource, not Azure OpenAI.
Content Safety text and image moderation
Azure AI Content Safety analysis that scores text or images for hate, sexual, violence and self-harm by severity; it doesn't detect prompt injection.
Content Understanding agentic mode
Preview mode (2026-06-01-preview) that reasons over document evidence for calculations and validation, initially one file per request.
Content Understanding analyzer (analyzer)
Configuration in Content Understanding that defines how content is processed and which fields are extracted, via prebuilt analyzers or a custom field schema.
Content Understanding pro mode
Preview mode, available only in the now-retired 2025-05-01-preview API, adding reasoning, multiple input files and a reference knowledge base for cross-document validation; being replaced by agentic mode.
Content Understanding service limits
Content Understanding input limits; documents accept PDF, TIFF, Office files, JPEG, PNG, BMP and HEIF/HEIC but not WebP.
Content Understanding standard mode
Default Content Understanding processing mode with the lowest cost and latency, one file per request; suits high-volume single invoices.
ContentSafetyClient
Content Safety SDK client, created with the endpoint and a key or Entra credential, whose instance methods AnalyzeText and AnalyzeImage (Python analyze_text, analyze_image) run moderation.
Context window
Total tokens (input plus output) a model can process in one request; for example GPT-4.1-mini supports about 1M tokens, Phi-3-mini (now retired) 128k.
Contract model (prebuilt-contract)
Document Intelligence prebuilt model that extracts contract fields such as Parties, Jurisdictions, Contract ID and Title.
Conversation (Foundry Agent Service) (conversation object)
Durable server-side object with an ID that stores items (messages, tool calls, tool outputs) across turns and sessions; a response run against it gets the full history and appends its output.
Cosine similarity (cosine distance)
Measure of the angle between two embedding vectors used to score semantic similarity in vector search; only embeddings models produce the vectors, not chat models.
Custom analyzer (Content Understanding)
Analyzer you define by extending a base analyzer such as prebuilt-document with your own field schema, methods and validation instructions.
Custom categories (Content Safety)
Content Safety feature for detecting your own content categories, either trained (standard, hours of training, retired on 1 September 2026) or rapid from samples; more effort than a blocklist for niche terms.
Custom classification model (custom classifier)
Document Intelligence model trained on at least two document types to classify and split files by type before extraction; confidence thresholds are set on its output.
Custom extraction model
Document Intelligence model you label and train to extract your own fields; needs as few as five labelled samples and comes as custom template or custom neural.
Custom keys connection
Project connection type that stores arbitrary key-value secrets, such as an API key named to match an OpenAPI securitySchemes entry, so the service injects the header at run time.
Custom NER (custom named entity recognition)
Azure Language feature you train to recognise domain-specific entities; it finds but doesn't mask them.
Custom neural model
Deep-learning custom extraction model that generalises across varied layouts of one document type (structured and semi-structured); the recommended starting point over custom template.
Custom speech
Speech feature that fine-tunes a base model with your text or audio plus transcripts for domain terms; create a project, pick a base model, upload data, train, test, then deploy.
Custom speech endpoint
Deployment that hosts a custom speech model for real-time recognition (endpointId); batch transcription can use the model without one, and you can swap models with no downtime.
Custom speech model expiration
Lifecycle rule: after expiry, endpoint requests fall back to the latest base model for the locale, while batch requests naming the expired model fail with 4xx; the model isn't deleted.
Custom subdomain (Foundry Tools) (custom domain name)
Resource-specific endpoint https://<name>.cognitiveservices.azure.com required for Microsoft Entra ID authentication; regional endpoints don't accept Entra tokens and the change can't be reversed.
Custom template model (custom form)
Document Intelligence custom extraction model for one fixed visual layout (1-5 minutes training); each layout variation needs its own model.
Custom text classification
Azure Language feature you train to assign whole documents to classes you define, such as work or personal email.
Custom Vision (Azure AI Custom Vision)
Foundry Tool for training your own image classification and object detection models from labelled images; it extracts no document fields and is retiring (support until 25 September 2028).
Custom voice endpoint (deploymentId endpoint)
Endpoint (voice.speech.microsoft.com/cognitiveservices/v1?deploymentId=...) that synthesises speech with a deployed custom voice; it doesn't list voices.
Custom Web API skill (WebApiSkill)
Skill (Microsoft.Skills.Custom.WebApiSkill) that calls your HTTP endpoint via uri, such as an Azure Function wrapping the Document Intelligence invoice model; a context ending /* runs it once per item.
Customer-managed keys (Azure AI Search) (CMK)
Optional second encryption layer over indexes and synonym maps using your RSA key in Azure Key Vault or Managed HSM, reached through the service's managed identity; it adds no capacity and can slow queries by 30-60%.
D
DALL-E (dall-e-3)
Retired OpenAI image generation model; dall-e-3 was retired in Azure on 4 March 2026 and replaced by the gpt-image series.
DefaultAzureCredential
Azure SDK credential that tries a chain of sources, using the signed-in developer locally and the managed identity on an Azure host, so no secret or key is stored.
Description (Image Analysis 3.2) (Describe, VisualFeatureTypes.Description)
Image Analysis 3.2 visual feature that returns one or more human-readable sentence captions ranked by confidence (read Description.Captions[0].Text); Caption replaces it in v4.0.
Document attack (indirect attack, indirect prompt injection)
Hidden instructions in third-party content (documents, emails, web pages, OCR text from images) that try to hijack the model session; caught by Prompt Shields for documents.
Document cracking
First indexer stage, which opens source files or records and extracts text, metadata and optionally images.
Document Extraction skill (DocumentExtractionSkill)
Utility skill that cracks a file passed in the enrichment pipeline (file_data) into text and optional images; it doesn't read pixels.
Document Intelligence add-on capabilities (features)
Optional features enabled through the features query parameter (keyValuePairs, queryFields, barcodes, formulas, styleFont, ocrHighResolution, languages).
Document Intelligence service limits
Document Intelligence input limits: 500 MB per file on S0 (4 MB on F0), images 50 × 50 to 10,000 × 10,000 pixels, and custom neural training data up to 1 GB.
Document Intelligence Studio
Web tool for trying Document Intelligence models and for labelling, training and composing custom template, neural and classification models on a Blob container.
Document translation (Azure.AI.Translation.Document)
Asynchronous Translator batch feature that translates whole documents between source and target blob containers while keeping formatting, with optional glossaries.
Document translation glossary (glossaryUrl)
Custom term list blob, placed in the target container and referenced by glossaryUrl, that forces specific translations; ignored if its language pair doesn't match.
Document-level access control (native ACL)
Preview Azure AI Search feature that indexes ADLS Gen2 or SharePoint permissions and trims results using the caller's Entra token passed with the query.
Domain-specific content detection (celebrities and landmarks models)
Image Analysis 3.2 feature (analyzeImageByDomain, or the details parameter) that recognises celebrities or landmarks; it returns names, not a sentence caption.
E
effort (Claude) (output_config effort)
Claude parameter that trades thoroughness for speed and cost; high (or xhigh, max) gives the deepest reasoning, and low or medium are cheaper and faster.
Embeddings (vector embeddings)
Numeric vectors that represent the meaning of text so that similar content sits close together (compared by cosine similarity); changing the vector length or model means re-embedding everything.
Embeddings model (text-embedding-3)
Model that converts text into vectors for semantic search, similarity and recommendations; it doesn't generate text.
enable_content_recording
Tracer setting that controls whether message content and tool arguments are recorded in traces; set it to False to redact them.
enableSegment (contentCategories)
Content Understanding analyzer setting that splits a file into segments and classifies each against contentCategories, optionally routing each to another analyzer.
Entity linking
Azure Language feature that disambiguates entities and links them to Wikipedia; it doesn't redact and retires on 1 September 2028.
Entity Linking skill (V3) (EntityLinkingSkill)
Built-in skill that returns recognised entities with links to Wikipedia articles.
Entity Recognition skill (V3) (EntityRecognitionSkill)
Built-in skill using Azure Language NER to extract entities in categories such as Person, Location, Organization and DateTime as metadata, not vectors.
estimateFieldSourceAndConfidence (estimateSourceAndConfidence)
Opt-in Content Understanding setting (analyzer-wide, or per field as estimateSourceAndConfidence) that returns a 0-1 confidence and source location for each field so low values can go to human review.
Exponential backoff
Retry pattern that doubles the wait after each failure, adds random jitter and caps retries; the recommended response to 429 and transient 5xx errors.
F
F0 pricing tier (Foundry Tools) (free tier)
Free SKU name for a Foundry Tools resource, passed as sku F0; limited transactions, and for example image captioning is free on ComputerVision F0.
Fabrication (hallucination)
Plausible but incorrect output that an LLM generates because it doesn't fact-check; grounding reduces but never eliminates it.
Fairness principle
Responsible AI principle that AI treats everyone fairly and similarly situated groups aren't affected differently; it is not the same as overall accuracy or identical output for all.
Few-shot learning (few-shot prompting)
Including example input-output pairs in the prompt to show the expected format and pattern for the current request; the model's weights don't change (zero-shot uses no examples).
Field generation method (extract, classify, generate)
Per-field method in a Content Understanding schema: extract (values as written, documents only), classify (pick from categories) or generate (infer or summarise).
Field mappings (fieldMappings)
Indexer setting that sends a source field verbatim to a differently named or typed index field; runs after document cracking and before the skillset.
Field schema (fieldSchema)
Content Understanding definition of the fields to extract, giving each a name, description, type and method (extract, classify or generate).
Figures (Document Intelligence) (output=figures)
Layout model output of detected charts and images; output=figures also generates cropped figure images you can download.
File projection
Knowledge store projection that saves only normalized binary images (/document/normalized_images/*) into a Blob container.
File search tool (file_search)
Agent tool that parses, chunks (800 tokens) and embeds uploaded files into a vector store and runs hybrid search over them; for files users upload, not an existing curated index.
Fine-tuning
Further training a pretrained model on a task-specific dataset to adjust its weights for style, format or task performance; not a safety control and not how to add fresh knowledge.
finish_reason
Chat completion field: stop means a normal finish, length means output was cut off by max_tokens or the token limit, content_filter means content was omitted, tool_calls means a tool was called.
Foundry Agent Service
Managed Microsoft Foundry platform for building, deploying and scaling prompt, voice and hosted agents with models, tools and guardrails.
Foundry Control Plane
Microsoft Foundry capability for centralised governance, monitoring and tracing of agents, including custom agents registered through an AI gateway into a Foundry project with Application Insights.
Foundry guardrail
Named collection of controls (risk, intervention point, action) in Microsoft Foundry; a guardrail assigned to an agent fully overrides the guardrail of its underlying model deployment.
Foundry IQ
Managed knowledge layer of knowledge bases and knowledge sources, built on Azure AI Search agentic retrieval, that gives Foundry agents permission-aware, cited grounding.
Foundry playground (model playground)
Foundry portal environment for testing a model or deployment interactively, adjusting system prompt, temperature, top_p and max tokens, comparing models and exporting sample code; not for production traffic.
Foundry project (project)
Child of a Foundry resource that isolates a team's agents, files, evaluations and project connections while sharing the resource's model deployments (resource-level connections can also be shared with every project).
Foundry project endpoint (project endpoint)
Project URL of the form https://<resource>.services.ai.azure.com/api/projects/<project> used to create AIProjectClient; it replaced the connection strings of early previews.
Foundry SDK (azure-ai-projects)
Microsoft Foundry client library that works with a project through AIProjectClient (agents, connections, deployments, datasets) and hands out an OpenAI-compatible client for responses, conversations and evaluations.
Foundry Tools (Azure AI services, Azure Cognitive Services)
Family of prebuilt and customisable AI APIs (Language, Speech, Vision, Content Understanding, Document Intelligence, Content Safety and more).
Foundry User (Azure AI User)
Least-privilege built-in developer role granting reader access plus data actions to build and test in a Foundry project; also assigned to a project's managed identity.
Foundry workflows (workflow)
Foundry (preview) orchestration that runs agents and logic as a declarative, deterministic sequence with if/else branching, variables and human-in-the-loop nodes; Foundry retires workflows on 1 December 2026 (the visual designer and in-portal execution stop being supported), with Agent Framework as the path.
Frequency penalty (frequency_penalty)
Inference parameter (-2.0 to 2.0) that penalises tokens by how often they've appeared, mainly reducing verbatim repetition.
Full-text search (keyword search)
Azure AI Search keyword retrieval over an inverted index ranked with BM25; best for exact identifiers, product codes and names, but misses paraphrases.
Function tool (function calling)
Agent or model tool you implement in your own code, declared in the tools array with type function, a name, a description and a JSON Schema of parameters; the app runs it and returns the output.
G
General document model (prebuilt-document)
Deprecated Document Intelligence model that extracted key-value pairs; replaced in v4.0 by the layout model with the keyValuePairs feature.
get_openai_client()
AIProjectClient method returning an authenticated OpenAI client for the project, used for responses.create (read response.output_text), conversations, evaluations and fine-tuning.
GitHub Actions
GitHub workflow automation triggered by events such as push or pull request, used to deploy Bicep or CLI templates and run Azure Machine Learning jobs without a person starting them.
Global Provisioned (GlobalProvisionedManaged)
Deployment type with reserved PTU capacity whose inference data may be processed in any Azure region; unlike Standard it doesn't keep processing in the resource's geography.
Global Standard (GlobalStandard)
Pay-per-token deployment type routed across Azure's global infrastructure with the highest default quota; the recommended starting point.
GPT (Generative Pretrained Transformer)
OpenAI's family of transformer LLMs pretrained on large text corpora for understanding and generating language and code.
GPT-3.5 Turbo
Earlier text-in, text-out OpenAI chat model; it can't interpret images or transcribe speech.
GPT-4 Turbo
Older, costlier GPT-4 chat model now retired in Azure; overkill for narrow text tasks that a small language model handles more cheaply.
GPT-4 vision-preview (GPT-4 Turbo with Vision preview)
Retired preview GPT-4 vision model (with OCR, object grounding and video enhancements) replaced by gpt-4 turbo-2024-04-09 and then GPT-4o; don't pick it for new image work.
GPT-4.1
OpenAI chat model series (gpt-4.1, mini, nano) with text and image input and text output; now deprecated in Azure (gpt-4.1-nano retires on 14 October 2026, gpt-4.1 and mini on 14 April 2027).
GPT-4o
Multimodal OpenAI chat model that accepts text and images (and audio in some variants); supports temperature and penalty parameters.
gpt-image-1 (gpt-image series)
OpenAI image generation model series that creates images from text prompts; it replaced DALL-E in Azure. The original gpt-image-1 (preview) itself retires on 23 October 2026, so new work uses gpt-image-1-mini or gpt-image-2 (gpt-image-1.5 also retires, on 16 December 2026).
gpt-image-1-mini
Cost-efficient, faster gpt-image model that supports inpainting with a mask but not input_fidelity or dedicated face preservation.
Ground truth (expected response)
Reference answer supplied with evaluation data that evaluators such as Response Completeness or F1 compare a model response against.
Groundedness evaluator
RAG evaluator that measures whether a response stays consistent with the provided context without fabricating content (the precision aspect).
Groundedness Pro evaluator
RAG evaluator (preview) that uses Azure AI Content Safety to return a strict true or false on whether a response is consistent with the retrieved context, with no model deployment needed.
Grounding (grounding data)
Supplying relevant, current facts in the prompt so the model answers from that context, reducing fabrication and enabling citations; the core of RAG.
Grounding data
Trusted content supplied in the prompt at request time so the model answers from it rather than guessing; it reduces fabrication.
Grounding with Bing Search (bing_grounding)
Agent tool, attached through a Bing project connection, that grounds answers in current public web results with citations; queries leave the Azure compliance boundary.
Group chat (workflow pattern) (group chat)
Multi-agent orchestration pattern in which control passes between agents dynamically, unlike a deterministic sequential workflow.
GStreamer
Open-source media framework the Speech SDK uses to decompress compressed audio input to PCM; it must be installed separately and be on the path.
Guardrail action (Block, Annotate)
Guardrail control setting: Block filters the detected content, while Annotate only labels and logs it without stopping it.
Guardrail intervention point
Point where a Foundry guardrail control scans content: user input, tool call (preview, agents only), tool response (preview, agents only) or output.
Guardrails + controls (Foundry portal) (Try it out)
Foundry portal area whose Try it out page runs the same Content Safety tests (moderate text or image content, protected material) against a Foundry resource.
H
haltOnBlocklistHit
Analyze text request flag; true skips further harm-category analysis once a blocklist term matches, false runs all analyses regardless.
Harm categories
The four classified content risks, hate and fairness, sexual, violence and self-harm, each scored safe, low, medium or high for text and images.
Hate and unfairness evaluator
Safety evaluator that detects hateful or unfair content in model or agent outputs; one of the risk and safety evaluators alongside violence, sexual, self-harm and protected material.
Health insurance card model (prebuilt-healthInsuranceCard.us)
Document Intelligence prebuilt model that extracts insurer, member, group and prescription fields from US health insurance cards, not medical reports.
Hosted agent
Foundry agent you build in your own code and framework, which Foundry runs as a container with a managed endpoint; contrast with a declarative prompt agent.
HTTP 403 Forbidden
Response when authentication succeeded but the identity lacks the required role or the network blocks it, for example az login without a data-plane role assignment.
HTTP 429 (Azure OpenAI) (Too Many Requests)
Response when a deployment's TPM or RPM limit is exceeded (or capacity is throttled); retry with exponential backoff and honour retry-after-ms, since failed requests still count.
HTTP 503 (Azure AI Search) (Service Unavailable)
Throttling response when a search service is at capacity; add replicas for query throttling, partitions for indexing throttling, or move to a higher tier.
Human in the loop
Design where a person reviews or approves AI actions; in a Foundry workflow an Ask a question node pauses for the reply and the next step is gated on the stored answer (for example approval == "approved").
Hybrid search
Single Azure AI Search query that runs full-text and vector search in parallel and merges results with Reciprocal Rank Fusion; with semantic ranker it gives the best overall relevance.
I
ID document model (prebuilt-idDocument)
Document Intelligence prebuilt model that extracts fields such as name, date of birth, document number and expiry from driver licences and passports.
Image Analysis
Azure Vision API that extracts visual features from an image in one call, such as a caption, Read OCR text, tags and detected objects with bounding boxes.
Image Analysis 3.2
Older Image Analysis version with the widest feature set: tags, objects, Describe captions, brands, faces, image type, colour scheme, landmarks, celebrities and adult content.
Image Analysis 4.0
Newer Image Analysis version with better models for Read, Caption, dense captions, tags, objects, people and smart crop; deprecated and retiring 25 September 2028.
Image Analysis skill (ImageAnalysisSkill)
Built-in skill returning image tags, captions, objects and faces from normalized images; it doesn't read text (use the OCR skill).
Image captioning
Azure Vision feature that generates a one-sentence, human-readable description of an image (dense captions describe up to 10 regions).
Image edit API (images/edits)
Images endpoint that edits an existing image from a prompt, optionally with a mask and input_fidelity; text-to-image generation ignores the original.
Image edit mask (mask)
PNG the same size as the input image whose fully transparent (alpha 0) pixels mark the only area the model may change.
Image generation tool (image_generation)
Built-in Foundry Agent Service tool that generates images from text prompts using a gpt-image-1 deployment plus an orchestrator model; it's a tool, not an input content type.
Image moderation (Content Safety) (Analyze Image API)
Content Safety image:analyze call that classifies hate, sexual, violence and self-harm with severity 0, 2, 4 or 6 and needs no training; it doesn't detect instructions hidden in images.
Image tagging (tags)
Image Analysis feature that returns one-word tags for objects, living things, scenery and actions, each with a confidence score.
Image type detection (ImageType)
Image Analysis 3.2-only feature that rates clip art on a 0–3 scale and flags line drawings; object detection and descriptions don't classify image type.
imageAction (generateNormalizedImages)
Indexer configuration parameter; generateNormalizedImages (with dataToExtract: contentAndMetadata) extracts and normalises images into /document/normalized_images for OCR and image analysis.
Images API (image generation API)
Azure OpenAI endpoint (images/generations) that creates images from prompts with gpt-image models; use it directly for editing and masks.
Immersive Reader (Azure AI Immersive Reader)
Foundry Tool that improves reading comprehension by isolating text, reading it aloud, translating and highlighting parts of speech.
Inclusiveness principle
Responsible AI principle that AI empowers and engages everyone, removing barriers through accessibility, language support and screen-reader or assistive technology.
Index field attributes (searchable, filterable, facetable, sortable, retrievable, key)
Per-field Azure AI Search settings: searchable (full text), retrievable (returned), filterable (exact-match \$filter), sortable, facetable (counts) and key (unique document ID).
Indexer (search indexer)
Azure AI Search crawler (pull model) that reads one data source, optionally runs a skillset and writes to one index, on demand or on a schedule as often as every five minutes.
Indexer schedule
Recurring run interval (minimum five minutes) that lets a long indexing job resume where it stopped when change detection is enabled; it doesn't add parallelism.
Indirect attack evaluator (XPIA (cross-domain prompt injected attack))
Risk and safety evaluator measuring whether a response fell for a jailbreak injected through retrieved documents or other context.
Inpainting
Editing only part of an image: the edit API takes the original plus a mask, and only the masked area is regenerated, preserving composition and lighting elsewhere.
input_fidelity
GPT-image edit parameter (low default, high) that preserves the style and features of input images, such as faces or product identity, without changing unrelated areas; not supported on gpt-image-1-mini.
Integrated vectorization
Azure AI Search feature in which an indexer chunks content (Text Split skill) and calls an embedding skill such as Azure OpenAI Embedding, with a matching vectorizer on the index for query-time vectorisation.
Intent recognition (Speech) (IntentRecognizer)
Former Speech SDK feature that mapped utterances to intents; retired 30 September 2025, so use speech to text followed by CLU or an Azure OpenAI model.
Invoice model (prebuilt-invoice)
Document Intelligence prebuilt model that extracts invoice fields such as InvoiceId, dates, vendor, customer, line items and totals.
IP firewall rules (Foundry Tools) (IP network rules)
Rules that admit specific public IP ranges to a Foundry Tools or Search public endpoint; traffic still crosses the internet, so it's the least secure isolation option.
J
JSON (JavaScript Object Notation)
Text data format for ARM templates and Cosmos DB documents.
JSONL (JSON Lines)
Format with one JSON object per line, required for fine-tuning training data (chat format, UTF-8 with BOM, under 512 MB) and batch files; not used for normal REST calls.
K
Key phrase extraction
Azure Language feature returning a plain list of the main concepts in text with no categories or confidence scores.
Key Phrase Extraction skill (KeyPhraseExtractionSkill)
Built-in skill that returns the key phrases of text as metadata.
Keyframe (key frame)
Representative frame sampled from each video shot that Content Understanding uses as visual input for field extraction.
Keyless authentication (Microsoft Entra ID authentication)
Recommended way to call Foundry models with a Microsoft Entra ID token and an RBAC role (such as Foundry User on the project) instead of an API key.
Keys and Endpoint
Resource Management page of a Foundry Tools or Azure OpenAI resource in the Azure portal that shows the REST endpoint and Key1 and Key2.
KeywordRecognizer (keyword spotting)
Speech SDK class that listens locally for a custom keyword (.table model) and raises an event when it's heard; not general transcription.
Knowledge source
Agentic retrieval object defining indexed or remote content (Blob, OneLake, SharePoint, web) for a knowledge base; unrelated to knowledge stores.
Knowledge store (knowledgeStore)
Skillset property that persists enriched output to Azure Storage (Table storage and Blob only) through projections for analysis outside search; unrelated to Foundry IQ knowledge sources.
L
Language detection
Azure Language feature that identifies the language a text is written in; it doesn't translate or filter.
Language Detection skill (LanguageDetectionSkill)
Built-in skill that returns the predominant language code of each document, useful for filters and for downstream skills.
Language identification (LID)
Speech feature that identifies which of up to 4 (at-start) or 10 (continuous) candidate languages is spoken in audio, used with speech to text or speech translation; it doesn't filter profanity.
Language Studio
Legacy web portal for trying and customising Azure Language features; not a model catalogue, and Language features are now used in the Foundry portal.
Layout model (prebuilt-layout)
Document Intelligence model that extracts text, tables, selection marks and document structure, plus key-value pairs with the keyValuePairs feature.
LLM (large language model)
Neural network (usually a transformer) with billions of parameters trained on massive text to predict the next token; broadly capable but costlier and slower than an SLM.
Log Analytics workspace
Store for logs queried with KQL; used by Sentinel, VM insights and workspace-based Application Insights.
loggingOptOut
Azure Language query parameter that stops input being stored for up to 48 hours for troubleshooting; false by default for sentiment, key phrases, language detection and NER, true for PII and health.
Logit bias (logit_bias)
Parameter mapping token IDs to a bias from -100 to 100 that raises or lowers the likelihood of those specific tokens (-100 effectively bans one).
Long Audio API
Older asynchronous API for synthesising audio over 10 minutes, replaced by the Batch synthesis API and retiring on 1 April 2027.
LUIS (Language Understanding)
Legacy intent service replaced by CLU and now fully retired; it plays no part in building or using a custom voice.
M
Managed HSM
Single-tenant, FIPS-validated HSM pool in Azure Key Vault that can hold the RSA key used for customer-managed key encryption.
Managed identity
Entra identity for Azure resources with no stored secret; system-assigned or user-assigned.
Markdown
Lightweight text format Content Understanding uses to represent extracted content (text, tables, transcripts) alongside fields.
Markdown output (Document Intelligence) (outputContentFormat=markdown)
Layout model option that returns content as Markdown with tables in HTML, useful for RAG chunking.
max_tokens (max completion tokens, max response)
Parameter that caps how many tokens the model generates in a response; input plus output must fit the model's context length.
MCP (Model Context Protocol)
Open protocol for exposing tools to models; the Foundry MCP tool connects an agent to a remote MCP server identified by a unique server_label and server_url, and tool_choice for it names that label.
Memory search tool (MemorySearchTool)
Agent tool that reads and writes a memory store; scope "{{\$userId}}" isolates memory per user, resolved from the x-memory-user-id header or else the caller's Entra tenant and object IDs.
Memory store (agent memory, long-term memory)
Managed long-term memory for Foundry agents (preview) that keeps distilled facts such as user preferences across sessions, partitioned by scope; not a full conversation history.
Microsoft Agent Framework (Agent Framework)
Open-source SDK, successor to Semantic Kernel and AutoGen, for building agents and graph-based multi-agent workflows with tools, memory, session state and human-in-the-loop; the recommended orchestration layer for hosted Foundry agents.
Microsoft Defender for Cloud Apps (Microsoft Cloud App Security)
Microsoft CASB for SaaS app discovery, session control and data protection; integrating it with Defender for Cloud unlocks no Defender for Cloud features.
Microsoft Entra Agent ID
Identity and security framework that gives AI agents their own Entra identities (agent identities created from blueprints) so Conditional Access, ID Protection and governance apply to them as to users.
Microsoft Entra ID (Azure AD)
Microsoft's cloud identity service and tenant for Azure and Microsoft 365.
Microsoft Fabric
SaaS analytics platform built on OneLake; Foundry agents reach its enterprise data through the Microsoft Fabric data agent tool.
Microsoft Fabric data agent tool (Fabric tool)
Agent tool that queries a Fabric data agent over enterprise analytics data through a project connection, using the signed-in user's identity (on-behalf-of).
Microsoft Foundry (Azure AI Foundry)
Azure platform for building, evaluating and running AI agents and models under one resource, with guardrails, evaluations and AI red teaming.
Microsoft Foundry AI agent evaluation GitHub Action (ai-agent-evals)
GitHub Action (preview) that runs offline evaluations in CI/CD: it sends a test dataset to Foundry agents, scores responses with catalogue evaluators, compares versions statistically and can gate a release.
Microsoft Foundry project (Foundry project)
Development boundary inside a Foundry resource for building and evaluating generative AI agents and apps; not a tool for governing traditional ML assets.
Microsoft Foundry resource (Foundry resource)
Top-level Azure resource (Microsoft.CognitiveServices) that holds model deployments, security and networking settings and child projects; required for Content Understanding, unlike a standalone Azure OpenAI resource.
Microsoft Responsible AI Standard
Microsoft framework for building AI systems on six principles: fairness, reliability and safety, privacy and security, inclusiveness, transparency and accountability.
Microsoft.CognitiveServices (Microsoft.CognitiveServices/accounts)
Azure resource provider for every Foundry, Azure OpenAI and Foundry Tools account; its ARM path sits under subscriptions/{id}/resourceGroups/{rg}/providers and never contains the tenant ID.
Microsoft.MachineLearningServices
Azure resource provider for Azure Machine Learning workspaces, not for Foundry Tools accounts.
Microsoft.Search
Azure resource provider for Azure AI Search services (searchServices); there is no Microsoft.CognitiveSearch provider.
Model deployment (deployment)
A named instance of a Foundry model in a resource, with a deployment type and TPM quota; the deployment name is what requests pass as the model.
Model deployment name
Name you give a deployment and pass in the model parameter; requests route by deployment name (for example my-mini-gpt), not the underlying model name.
Model router
Deployable Foundry model that routes each prompt in real time to a suitable underlying model (small and cheap for simple prompts, reasoning or frontier for complex ones) behind one deployment, with Balanced, Cost or Quality routing modes.
Model version upgrade policy
Deployment setting (auto-update to default, upgrade when expired, or no auto upgrade) that controls when a Standard deployment moves to a new model version.
Moderate text content (Content Safety Studio)
Studio and Foundry Try it out feature that tests single sentences or bulk datasets, shows category and severity, adjusts filter thresholds and blocklists, and exports code.
Monitor online activity (Content Safety Studio)
Studio page that reports past moderation API usage and trends, such as category and severity distribution, block rate, latency and blocklist hits; it tests nothing new.
mstts:express-as
Microsoft SSML extension that sets a voice's speaking style (for example calm, gentle or advertisement_upbeat); an invalid style makes the whole element be ignored.
mstts:express-as role
Mstts:express-as attribute that makes a voice role-play a different age and gender (such as YoungAdultFemale or SeniorMale) without changing the voice name; unsupported roles are ignored.
Multi-service Foundry resource (Microsoft Foundry resource)
Single Azure resource giving one endpoint and key for Azure OpenAI and multiple Foundry Tools, as opposed to a single-service resource for one tool.
Multimodal model (vision-enabled model)
Generative model that accepts more than one input type, such as text plus images, in a single request.
N
NER (named entity recognition)
Azure Language feature that finds entities and labels them with categories such as Person, PersonType, Location, Organization, DateTime, Quantity and Skill.
Neural voice (standard voice)
Prebuilt deep-neural-network text to speech voice available out of the box in 100+ languages and locales.
Normalized images (normalized_images)
Enrichment-tree node (/document/normalized_images/*) of resized, rotated images produced during document cracking; the only input OCR, Image Analysis and file projections accept.
O
Object detection
Computer vision technique that returns a label, confidence score and bounding-box coordinates for each object instance, so you can locate and count multiple objects or classes.
Object projection
Knowledge store projection of whole JSON enriched documents into a Blob container (storageContainer, source), for data science.
Ocp-Apim-Subscription-Key (Foundry Tools)
HTTP header that carries a Foundry Tools resource key (for example Language or Translator); Azure OpenAI REST calls use the api-key header instead.
Ocp-Apim-Subscription-Region
Translator request header naming the resource region; required with multi-service or regional keys and optional for a global Translator resource.
OCR (optical character recognition)
Extracting printed or handwritten text from images and documents so it can be analysed; Azure Language can't read scanned images itself.
OCR skill (OcrSkill)
Built-in skill using Azure Vision Read to extract printed and handwritten text from /document/normalized_images/*; needs imageAction set on the indexer.
OData filter (\$filter)
Azure AI Search query parameter using OData syntax against filterable fields for exact-match filtering.
OpenAPI tool (OpenApiTool)
Agent tool that calls an external REST API described by an OpenAPI 3 spec; for API-key auth the spec declares an apiKey securitySchemes entry and the tool is bound to a project connection holding the key.
OpenID Connect
Identity layer on OAuth 2.0 used for sign-in and for workload identity federation, letting GitHub Actions get Azure tokens without a stored secret.
OpenTelemetry (OTel)
Open standard for traces, metrics and logs that Foundry tracing and Azure Monitor Application Insights use, including GenAI semantic conventions.
Operation-Location
Response header returned by asynchronous Content Understanding and Document Intelligence calls (201 on analyzer PUT, 202 Accepted on analyze); poll this URL until the status is Succeeded.
Operation-Location header
Response header returned by asynchronous APIs such as Read 3.2 (ReadAsync) and batch document translation; it holds the operation ID you poll (GetReadResultAsync) until the status is succeeded.
Opinion mining (aspect-based sentiment analysis)
Sentiment analysis option that links sentiment to specific targets (aspects) and their assessments in the text.
OTEL_SERVICE_NAME
OpenTelemetry environment variable that sets the service.name resource attribute (the cloud role name), separating services that share one Application Insights resource.
Output field mappings (outputFieldMappings)
Indexer setting that maps enriched-document nodes created by skills to index fields; runs after the skillset and is required for any enriched content you want indexed.
output_text
Convenience property of a Responses API result that holds the model's generated text; output items are responses, not input content types.
outputType (Content Safety) (FourSeverityLevels, EightSeverityLevels)
Analyze text request field that returns severities as 0, 2, 4, 6 (FourSeverityLevels, the default) or 0 to 7 (EightSeverityLevels); images always use four levels.
P
Parallel indexing
Partitioning content into containers or virtual folders with one data source and indexer per partition, all targeting the same index and skillset, and running them together; limited to about one indexer job per search unit.
Past messages included (conversation history)
Playground setting for how many previous chat turns are sent with each request; more history gives context but uses more tokens.
Personal access token (GitHub) (PAT)
GitHub credential for GitHub APIs; it doesn't authenticate a workflow to Azure, which should use the Azure Login action with OpenID Connect.
Phi models (Phi-4, Phi-4-mini)
Microsoft family of small language models (for example Phi-4-mini-instruct, Phi-4-multimodal-instruct, Phi-4-reasoning) for narrow, low-cost tasks; newer Phi-4 models replace Phi-3.
Phi-3-mini (Phi-3-mini-128k-instruct)
Small Microsoft language model with at most a 128k-token context (4k variant also existed); now retired in favour of Phi-4-mini-instruct.
PII detection (personally identifiable information detection)
Azure Language prebuilt feature that detects entities such as Person, PhoneNumber, Email and Address and returns redacted (masked) text; not harmful-content moderation.
PII guardrail (personal data filter)
Foundry guardrail control that detects personal data such as names, emails and phone numbers in model output and annotates, redacts or blocks it per category.
Power BI
Microsoft reporting service; connects to on-premises data via the on-premises data gateway.
Power Fx
Excel-like low-code formula language used in Foundry workflows; variables need a scope prefix (Local. or System.), IsBlank tests a value, IsEmpty tests a table and Upper() capitalises text.
Prebuilt analyzer
Ready-to-use Content Understanding analyzer such as prebuilt-read, prebuilt-layout, prebuilt-invoice, prebuilt-audioSearch or prebuilt-callCenter; can be customised with extra fields.
prebuilt-documentFieldSchema
Content Understanding utility analyzer that proposes a field schema from a document.
prebuilt-documentSearch
Content Understanding RAG analyzer that extracts paragraphs, tables and figure descriptions as Markdown with a summary, using generative models.
Presence penalty (presence_penalty)
Inference parameter (-2.0 to 2.0) that penalises any token already present; positive values push the model to new topics.
Privacy and security principle
Responsible AI principle that personal and business data is protected, with notice, consent and restricted access; it is not about handling unusual inputs.
Private endpoint
Private IP for a service in your VNet, reachable from peered VNets and from on-premises over ExpressRoute or VPN; public access can then be disabled.
Professional voice fine-tuning (custom neural voice Pro)
Limited-access Azure Speech feature, built in Speech Studio or the Foundry portal, that trains a brand voice from a consenting voice talent's recordings: create the project, add the talent's consent recording, upload training audio, fix data issues, train and deploy.
Project connection (Foundry connection, connected resource)
Stored endpoint and auth (API key or keyless Entra ID) for an external resource such as Azure AI Search, Storage or Bing that every app and agent in the project shares, referenced by its connection ID (project_connection_id).
Projection group
One element of a knowledge store's projections array, holding related tables, objects and files; one element with several projections is still one group.
Prompt agent
Declaratively defined Foundry agent made of a model, instructions and tools that Foundry runs for you with no code or containers to manage.
Prompt engineering
Crafting prompts, system messages and examples to steer a model's output without changing its weights.
Prompt flow
Azure Machine Learning and Foundry tool for building LLM apps as a graph of nodes (LLM, Python, prompt tools) with variants and evaluation; retires 20 April 2027 in favour of Microsoft Agent Framework.
Prompt Shields
Content Safety and Foundry guardrail feature that detects adversarial inputs: user prompt attacks (jailbreaks) at user input and document attacks hidden in third-party content at user input and tool response.
Prompt Shields for documents (document attack, indirect attack, XPIA)
Prompt Shields control that catches instructions hidden in third-party content such as grounding data, emails, OCR text and tool responses.
Prompt Shields for user prompts (user prompt attack, jailbreak)
Prompt Shields control that catches direct attacks in the user's own input (rule changes, role-play personas, conversation mockups, encoding tricks); it doesn't scan documents or images.
PromptAgentDefinition
Foundry SDK definition of a prompt agent (model, instructions, tools such as a memory search tool) passed to agents.create_version.
Protected material detection
Content Safety filter that scans model output for known copyrighted text (lyrics, articles, recipes, selected web content) and code from public GitHub repositories; English only.
Protected material evaluator
Risk and safety evaluator that detects copyrighted or otherwise protected text (such as lyrics or articles) in outputs.
Provisioned throughput (PTU, provisioned throughput units)
Deployment type that reserves dedicated model capacity measured in PTUs and billed hourly whether used or not, for predictable throughput and latency.
Provisioned-managed Utilization (Provisioned Utilization)
Azure OpenAI metric of how much of a provisioned deployment's capacity is used; above 100% the deployment returns 429.
Push API (push model)
Loading JSON documents into an index programmatically (up to 1,000 documents or 16 MB per batch) from any source; the alternative for unsupported sources such as on-premises SQL Server, but it can't run skillsets.
R
RAG (retrieval augmented generation)
Pattern that retrieves relevant content from your data (often via an index), adds it to the prompt as grounding data and generates a cited answer; used for private or recent information.
RBAC (role-based access control)
Azure role assignments governing who can manage resources, inherited down scopes; not where or what size resources are.
Read (OCR) (Read API)
Azure Vision OCR engine that extracts printed and handwritten text with locations and confidence scores; for text-heavy documents use Document Intelligence or Content Understanding.
Read model (prebuilt-read)
Document Intelligence OCR model that extracts printed and handwritten text, lines, words and languages only; recommended for text-heavy scans and the engine under the other models.
Real-time transcription
Speech to text mode that transcribes streaming audio (microphone or file) as it is recognised, returning intermediate results; used for live captions and voice input.
Receipt model (prebuilt-receipt)
Document Intelligence prebuilt model that extracts sales receipt fields such as MerchantName, transaction date, tax and total.
Reflection (self-critique)
Generation pattern where the model or an evaluator compares a draft with its sources, lists what is missing and revises before answering; the fix when retrieved content is omitted.
regenerateKey (az cognitiveservices account keys regenerate)
Management action (POST .../regenerateKey with keyName Key1 or Key2) that resets only the named key of a Foundry Tools account, so clients can move to the other key for zero-downtime rotation; it writes nothing to Key Vault.
Regional endpoint (Foundry Tools)
Shared per-region endpoint such as eastus.api.cognitive.microsoft.com; it works with keys but not with Entra ID authentication.
Relevance evaluator
RAG evaluator that measures how accurately and directly a response addresses the user's query.
Reliability and safety principle
Responsible AI principle that AI performs as intended across conditions, responds safely to unusual or missing input and resists manipulation; declining to predict on missing fields belongs here.
RequestResponse log
Azure OpenAI resource log category recording requests with status codes and latency; Audit logs cover administrative operations and Trace logs detailed inference traces.
Response compaction (responses.compact)
Responses API operation that shrinks a long conversation's context into compaction items so later turns fit the context window; responses.retrieve fetches an existing response.
Response Completeness evaluator
RAG evaluator (preview) that measures whether a response covers the critical information in the ground truth (the recall side, where groundedness is the precision side).
Responses API
Azure OpenAI and Foundry API that takes an input array of messages and content items (input_text, input_image) and returns output items, with the text in response.output_text.
REST (representational state transfer)
HTTP API style used by Azure services, e.g. to start a compliance scan.
ResultReason
Speech SDK result enum: RecognizedSpeech or TranslatedSpeech mean success, NoMatch means nothing was recognised, Recognizing/TranslatingSpeech mark intermediate results and Canceled signals an error.
retry-after-ms
Azure OpenAI response header on a 429 giving how long to wait before retrying; use exponential backoff with jitter when it's absent.
Risk and safety evaluators
Foundry evaluators for hate and unfairness, sexual, violence, self-harm, protected material, indirect attacks, prohibited actions, sensitive data leakage and more; they score outputs but don't block anything.
RPM (requests per minute)
Per-deployment request-rate limit derived from the TPM allocation and evaluated over short 1- or 10-second windows, so bursts can hit 429 below the per-minute total.
S
S0 pricing tier (Foundry Tools) (standard tier)
Standard paid SKU name (S0) for Foundry, Azure OpenAI and most Foundry Tools resources; Document Intelligence offers only F0 and S0, while Computer Vision's standard tier is S1.
Safety system message
System message that adds explicit boundaries and refusal guidance to mitigate harms; one layer of a safety stack, not a complete control.
Sample Labeling tool (Form Recognizer sample labeling tool)
Retired legacy tool for labelling Form Recognizer v2.1 training data; use Document Intelligence Studio or the Foundry portal.
Search analyzer (analyzer, indexAnalyzer, searchAnalyzer)
Text-processing component (standard Lucene, language, built-in or custom) that tokenises searchable string fields; it changes tokenisation, not meaning.
Search data source (data source)
Azure AI Search object holding the connection to supported content (Blob, ADLS Gen2, Cosmos DB, Table Storage, Azure SQL, SQL MI, SQL Server on Azure VMs); an indexer uses exactly one.
Search Index Data Contributor
Azure AI Search data-plane role that loads and modifies documents in indexes; the keyless replacement for admin-key data access.
Search Index Data Reader
Azure AI Search data-plane role for search, lookup, autocomplete and suggestions only, assignable at service or single-index scope; the keyless replacement for a query key.
Search unit (SU)
Billing and capacity unit of an Azure AI Search service (replicas × partitions); a service runs about one indexer job per search unit.
search.in
OData filter function for matching a field against a comma-separated list of values, e.g. group_ids/any(g:search.in(g, 'g1, g2')); faster than chained eq/or.
Security trimming (security filter)
Pattern that stores group or user IDs in a filterable field (such as group_ids) and filters each query to the caller's groups with search.in.
securitySchemes (OpenAPI) (apiKey security scheme)
OpenAPI spec section that, with type apiKey, in header and a name matching the connection key, plus a security section, tells the OpenAPI tool to inject the key; without it no header is sent and the API returns 401.
Selected Networks and Private Endpoints
Networking setting on a Foundry Tools resource that sets the default action to Deny and admits only listed virtual network rules, IP ranges and private endpoints.
Selection mark
Checkbox or radio button state (selected or unselected) extracted by the layout model and custom models.
Semantic Kernel (Semantic Kernel Agent Framework)
Earlier Microsoft SDK for orchestrating models, plugins and agents; superseded by Microsoft Agent Framework.
Semantic ranker (semantic ranking)
Azure AI Search feature that reranks the top BM25 or hybrid results with Bing-derived language models and can add captions and answers; the re-ranker to add without changing embeddings.
Sentiment analysis
Azure Language feature returning positive, neutral, negative or mixed labels with confidence scores per sentence and document.
Service endpoint
Routes a subnet's traffic to a service's public endpoint over the Azure backbone; free, but not usable from on-premises.
Service principal
App-registration identity with a stored secret or certificate that must be rotated and can be copied; suits code outside Azure.
Severity threshold
Guardrail setting per harm category (default medium) at or above which content is blocked; blocking at low on input and output is the strictest setting.
Shaper skill (ShaperSkill)
Utility skill that assembles existing enrichment nodes into one complex object, typically the source of knowledge store table projections.
SIEM (security information and event management)
Security log analytics and detection, e.g. Microsoft Sentinel.
Skillset
Reusable Azure AI Search object that lists the skills (built-in or custom) an indexer runs during enrichment, plus optional knowledge store projections and a billing resource.
SLM (small language model)
Language model with the same kind of architecture as an LLM but fewer parameters (roughly under 10 billion), so cheaper and faster but less broadly capable.
speak_text_async()
SpeechSynthesizer method that synthesises plain text without blocking and returns a future for the SpeechSynthesisResult; speak_ssml_async() takes SSML instead.
Speaker recognition (voice biometrics)
Azure Speech capability that identifies or verifies who is speaking from voice characteristics; a Limited Access feature, distinct from speech recognition, which transcribes what is said.
Speech SDK
Client library (Python package azure-cognitiveservices-speech, imported as speechsdk) exposing SpeechConfig, SpeechRecognizer, SpeechSynthesizer and audio config classes.
Speech Studio
Web portal of no-code tools for Azure Speech (custom speech, custom voice, audio content creation, batch speech to text); agents aren't created there.
Speech to text (speech recognition)
Azure Speech capability that converts spoken audio into text (captions, call and meeting transcripts, voice commands); it doesn't identify who is speaking.
Speech translation
Azure Speech capability that translates spoken audio in real time into text or synthesised speech in one or more target languages.
SpeechRecognizer
Speech SDK class for speech to text, built from SpeechConfig and an AudioConfig (microphone, WAV file or stream); use it when no translation is needed.
SpeechSynthesizer
Speech SDK class for text to speech (SpeakTextAsync, SpeakSsmlAsync) that outputs to a speaker, file or stream.
SpeechTranslationConfig
Speech SDK configuration for speech translation: SpeechRecognitionLanguage sets the source locale and AddTargetLanguage adds each target language code.
Spotlighting
Prompt Shields sub-feature (preview, off by default, Chat Completions only) that base64-encodes document content so the model treats it as lower trust than system and user prompts; it adds tokens.
SQL Server on Azure VMs
IaaS SQL Server with full OS control; HA through Always On AGs or FCIs.
SSML (Speech Synthesis Markup Language)
XML-based markup for text to speech that controls voice, pitch, rate, volume, pauses and pronunciation; sent with speak_ssml_async().
SSML emphasis element (emphasis)
SSML element that adds or removes word-level stress (reduced, none, moderate, strong); supported only by a few neural voices.
SSML prosody element (prosody)
SSML element that adjusts pitch, contour, range, rate and volume of synthesised speech.
SSML voice effect (eq_car, eq_telecomhp8k)
Optional effect attribute on the SSML voice element that optimises output for a playback scenario: eq_car for cars and enclosed vehicles, eq_telecomhp8k for 8 kHz telephony.
SSML voice element (voice name)
SSML element whose name attribute picks the voice for the enclosed text; the accent comes from the voice's locale, so two en-US voices share one accent.
Standard deployment type (Standard)
Pay-per-token Foundry deployment type that processes data within the resource's Azure geography, for geography compliance at lower volume.
Stop sequence (stop)
Up to four strings at which the model stops generating; the output ends before the sequence.
Storage Blob Data Contributor
Built-in data-plane role that reads, writes and deletes containers and blobs; with Reader, the least privilege for portal uploads.
Storage Blob Data Owner
Built-in data-plane role with full access to blob containers and data, including setting POSIX ACLs and blob index tags.
Storage Blob Data Reader
Built-in data-plane role that reads and lists containers and blobs; not accepted for Content Safety blob-URL access, which needs Storage Blob Data Contributor or Owner.
Streaming (stream=true)
Returning model output incrementally as server-sent events; it changes only how the answer is delivered, not its content, cost or completeness.
Structured outputs (response_format, text.format)
Feature that forces output to match a supplied JSON Schema (json_schema, strict true), set in response_format for Chat Completions or text.format for Responses; it controls format, not tool calls.
styledegree
Mstts:express-as attribute that sets style intensity from 0.01 to 2 (default 1; 2 doubles the intensity).
Suggester
Index construct listing sourceFields for autocomplete and suggestions; one per index, and source fields must use the standard or a language analyzer.
Synonym map
Service-level Azure AI Search resource of Solr-format rules assigned to searchable string fields to expand equivalent terms at query time.
System message (system prompt, metaprompt)
High-priority instructions and context sent first in a chat request to set the model's role, tone, format and boundaries; it steers but doesn't guarantee compliance.
System-assigned managed identity
Identity created and deleted with one resource; ten VMs get ten identities; Azure Policy remediation can use one.
T
Table projection
Knowledge store projection into Azure Table Storage rows, related by generated keys (tableName, generatedKeyName, source); the choice for Power BI.
Table storage
Cheap key-value tables indexed on PartitionKey and RowKey only, with one write region and 1 MB entities.
Temperature
Inference parameter (0 to 2) controlling randomness; low values such as 0.2 give focused, near-deterministic output and high values more creative output.
Text Merge skill (MergeSkill)
Utility skill that folds OCR text or image captions back into the document content (merged_content); it doesn't chunk.
Text Split skill (SplitSkill)
Utility skill (free) that breaks text into pages or sentences with optional overlap, producing chunks for embedding; it doesn't vectorise.
Text to speech (speech synthesis)
Azure Speech capability that converts text into humanlike spoken audio with standard or custom neural voices, tuned with SSML; the opposite direction to speech recognition.
Text Translation skill (TextTranslationSkill)
Built-in skill that translates text with Translator; omit defaultFromLanguageCode to auto-detect the source language.
text-embedding-ada-002 (ada-002)
Azure OpenAI embedding model that only returns 1,536-dimension vectors (8,192 max input tokens); it generates no text.
TextBlocklistMatch (blocklists_match)
Content Safety result type listing a blocklist hit (blocklist name, item ID and matched text), returned in blocklists_match; not a client or request.
TextCategoriesAnalysis (categories_analysis)
Content Safety result type giving one harm category and its severity; analyze_text returns a list of these in categories_analysis, and it is not a client or request.
TextTranslationClient (Azure.AI.Translation.Text)
Translator text SDK client created with AzureKeyCredential and region; TranslateAsync returns TranslatedTextItem results and GetSupportedLanguagesAsync lists languages.
textType (Translator) (textType=html)
Translate parameter: plain (default) treats tags as text, while html preserves well-formed markup and honours class=notranslate.
Token
Chunk of text (word, part word or punctuation) that an LLM processes; context windows, limits and billing are measured in tokens.
tool_choice
Request parameter that controls tool calling: auto (default, model decides), required (must call at least one tool), none (no tool calls) or a specific tool's specification to force that tool.
tool_resources (ToolResources)
Agent or conversation field that supplies the data a tool uses, such as vector store IDs for file search or files for code interpreter; tool definitions themselves go in tools.
Top P (nucleus sampling)
Inference parameter that limits token choice to the top probability mass, controlling diversity rather than length; adjust it or temperature, not both.
total_tokens (usage)
Usage field equal to prompt_tokens plus completion_tokens (e.g. 37 + 86 = 123); both prompt and completion tokens are billed.
TPM (tokens per minute)
Unit of Azure OpenAI quota and deployment rate limit; exceeding it returns HTTP 429.
Tracing (Foundry)
Foundry observability feature (off by default) that records each request as an OpenTelemetry trace with inputs, outputs, tool calls, token usage and latency, stored in the connected Application Insights resource.
TranslationRecognizer
Speech SDK class that translates spoken audio into text (and optionally synthesised speech) in target languages; success is ResultReason.TranslatedSpeech.
Translator custom endpoint (custom domain endpoint)
Resource endpoint <name>.cognitiveservices.azure.com/translator that must be used once Translator has network restrictions; the global and geography (api-nam) endpoints are refused.
Translator detect operation (/detect)
Translator operation that returns a text's language code and confidence score without translating it.
Translator geography endpoints (api-nam, api-apc, api-eur)
Text translation endpoints that keep processing in one geography (Americas, Asia Pacific, Europe), unlike the global endpoint, which can fail over outside it.
Translator languages operation (/languages, GetSupportedLanguages)
Translator operation (no authentication needed) that lists supported languages for translation, transliteration and dictionary scopes.
Translator translate operation (/translate)
Translator text operation that translates to one or more targets (repeat to=), auto-detects the source when from is omitted, and supports textType and toScript.
Transliteration (/transliterate, toScript)
Converting text from one script to another (for example Thai to Latn); /transliterate changes script only, while toScript on /translate adds a transliteration of the translation.
Transparency note
Microsoft document describing an AI feature's capabilities, limitations and appropriate uses; the Sentiment Analysis note warns against automatic actions such as employee bonuses based on sentiment scores.
Transparency principle
Responsible AI principle that people know when AI is used, what it can and can't do and how it reaches outputs, such as explaining a loan decision.
U
User message (user prompt)
Chat message role that carries the end user's question and inputs such as an image, as opposed to the system message's standing instructions.
User prompt attack (jailbreak, direct attack)
Malicious prompt from the user that tries to bypass the system message or safety training; Prompt Shields for user prompts catches it but not instructions hidden in documents.
V
Vector search
Similarity search over embeddings that finds content by meaning and paraphrase (nearest neighbours such as HNSW), but can miss exact identifiers.
Vector store
Agent Service container of chunked, embedded file content that the file search tool searches; up to 10,000 files, one per agent and one per conversation.
Virtual network rule (VNet rule)
Network rule on a resource (Foundry Tools, Storage and others) that admits a specific subnet, which must have the matching service endpoint (e.g. Microsoft.CognitiveServices) enabled; it takes effect only when the default action is Deny.
Vision-enabled chat model (large multimodal model)
Chat model (GPT-4o, GPT-4.1, GPT-5 series, o-series) that interprets images in the prompt; a text-only model can't be given this ability by the app layer.
Voice Live API (Voice Live)
Fully managed, low-latency speech-to-speech API for real-time voice agents that combines speech recognition, a generative model and text to speech in one WebSocket interface, with optional avatar and function calling; it returns audio, not only text.
Voice talent profile (voice talent)
Record in a professional voice project that holds only the voice talent's recorded consent statement, which is compared with the training audio to verify the speaker; training .zip files don't go here.
Voices list API
Text to speech REST GET call (/cognitiveservices/voices/list on the regional tts endpoint or /tts/cognitiveservices/voices/list on the resource endpoint) that returns standard voices with locale, gender and styles; it doesn't synthesise speech.
W
Whisper
OpenAI speech-to-text model for transcription and translation into English (25 MB file limit); available in Azure OpenAI and Azure Speech. The Azure OpenAI whisper model (001) retires on 15 December 2026.
Z
Zero-shot prompting (zero-shot)
Prompting with only the instruction and no examples; the prompted chat model can classify into categories listed at request time.