Appearance
Glossary
Definitions of key terms used throughout GateFlow documentation.
A
Agent
An AI system that can autonomously perform tasks using tools and resources. In GateFlow MCP, agents are governed entities with defined permissions, rate limits, and audit trails.
Agent ID
A unique identifier for an MCP agent, prefixed with agent_. Example: agent_abc123xyz.
API Key
A secret credential used to authenticate requests to the GateFlow API. Keys are prefixed by environment: gw_dev_ for development, gw_prod_ for production, gf-agent- for MCP agents.
Audit Log
An immutable, hash-chained record of all operations performed through GateFlow, used for compliance and debugging.
B
BAA (Business Associate Agreement)
A HIPAA-required contract between covered entities and business associates that handle PHI. GateFlow offers BAAs for healthcare customers.
Batch Processing
Processing multiple requests in a single API call, often with cost savings compared to individual requests.
Bates Numbering
A sequential numbering system used in legal discovery to identify and track documents. Format: PREFIX00000001.
C
Cache Hit
When a request matches a previously cached response, avoiding a call to the upstream provider.
Cache TTL (Time To Live)
How long a cached response remains valid before expiring.
Chain of Custody
A documented trail showing the handling of evidence or data, critical for legal compliance.
Chat Completion
The primary API endpoint for conversational AI, accepting messages and returning model responses.
Chunk
A segment of a document created during ingestion for vector storage and retrieval. Typical size: 500-1500 characters.
Classification (Data)
The sensitivity level assigned to data: public, internal, confidential, restricted, or phi.
Context Window
The maximum amount of text (measured in tokens) that a model can process in a single request.
Cross-Encoder
A neural network architecture used in reranking that processes query-document pairs together for high accuracy relevance scoring.
Custodian
In legal discovery, a person whose data is subject to a litigation hold.
D
Default-Deny
A security model where all permissions are denied unless explicitly granted. Used by GateFlow MCP.
Deprecation
The process of phasing out a model or API version. GateFlow provides deprecation alerts and migration assistance.
Document Ingestion
The process of uploading, processing, chunking, and indexing documents for retrieval.
E
Embedding
A numerical vector representation of text that captures semantic meaning. Used for similarity search.
Embedding Model
An AI model that converts text into embeddings. Examples: text-embedding-3-large, embed-english-v3.0.
EU AI Act
European Union regulation establishing requirements for AI systems based on risk levels.
F
Fallback
An alternative model or provider used when the primary choice is unavailable or fails.
FHIR (Fast Healthcare Interoperability Resources)
A standard for exchanging healthcare information electronically.
G
Gateway
The GateFlow service that routes requests to appropriate AI providers. Provides caching, fallbacks, and monitoring.
GDPR (General Data Protection Regulation)
EU regulation governing data protection and privacy, including the right to erasure (Article 17).
GPAI (General Purpose AI)
Under the EU AI Act, AI systems capable of performing a wide range of tasks.
H
HIPAA (Health Insurance Portability and Accountability Act)
US regulation governing the protection of health information (PHI).
HNSW (Hierarchical Navigable Small World)
An algorithm for approximate nearest neighbor search, used in GateFlow's vector database.
Human-in-the-Loop
A system design where human oversight is required for AI decisions, often mandated by the EU AI Act.
I
Intelligent Routing
GateFlow's system for automatically selecting the optimal model based on task requirements, cost, and latency.
J
JSON-RPC
The protocol used by MCP for communication between agents and the server.
L
Latency
The time between sending a request and receiving a response. Measured in milliseconds (ms).
Litigation Hold
A directive to preserve all data potentially relevant to pending or anticipated litigation.
LLM (Large Language Model)
An AI model trained on large amounts of text data, capable of generating human-like text. Examples: GPT-5.2, Claude Sonnet 4.5.
M
MCP (Model Context Protocol)
An open protocol for AI agent communication. GateFlow's MCP implementation adds governance, permissions, and audit logging.
Merged Cells
Table cells that span multiple rows or columns, requiring special handling during extraction.
Multi-Tenant
A system architecture where multiple customers share infrastructure while maintaining data isolation.
N
Native PDF
A PDF created digitally (not scanned) where text can be extracted without OCR.
O
OCR (Optical Character Recognition)
Technology that converts images of text into machine-readable text.
OpenAI SDK Compatible
APIs that work with the standard OpenAI client libraries by changing only the base_url.
Override
A human decision that supersedes an AI recommendation, tracked for audit purposes.
P
p50/p99 Latency
Percentile latency metrics. p50 is median latency; p99 is the latency experienced by 99% of requests.
PHI (Protected Health Information)
Health information that can identify an individual, protected under HIPAA.
PII (Personally Identifiable Information)
Data that can identify an individual, such as name, SSN, email, or address.
Pipeline
A sequence of processing steps. Voice pipelines combine STT → LLM → TTS.
Pipeline Template
A pre-configured pipeline for common use cases. Examples: voice-agent-fast, ambient-scribe.
Presidio
Microsoft's open-source PII detection engine, used by GateFlow for sensitive data identification.
Provider
An upstream AI service (OpenAI, Anthropic, Google, Mistral, Cohere, ElevenLabs) that GateFlow routes requests to.
Pseudonymization
Replacing identifying information with artificial identifiers while maintaining data utility.
R
RAG (Retrieval-Augmented Generation)
A technique that retrieves relevant context from a knowledge base and includes it in LLM prompts.
Rate Limit
A restriction on the number of requests allowed within a time period.
Reranking
A second-stage retrieval process that re-scores initial search results for higher relevance.
Residency (Data)
The geographic location where data is stored, important for regulatory compliance.
RLS (Row-Level Security)
Database security that restricts which rows a user can access based on their identity.
RPM (Requests Per Minute)
A rate limit metric measuring allowed requests per minute.
S
Scanned PDF
A PDF containing images of pages (not selectable text), requiring OCR for text extraction.
Semantic Cache
A cache that matches requests based on meaning similarity rather than exact string matching.
Semantic Search
Search based on meaning rather than keywords, using embeddings and vector similarity.
SLA (Service Level Agreement)
A commitment to uptime and performance, with credits for failures.
SOAP Note
A medical documentation format: Subjective, Objective, Assessment, Plan.
SSE (Server-Sent Events)
A protocol for servers to push updates to clients, used for streaming responses.
STT (Speech-to-Text)
Converting spoken audio to written text. Also called transcription or ASR.
stdio
Standard input/output, a transport method for MCP using command-line pipes.
T
Template (Pipeline)
See Pipeline Template.
Token
The basic unit of text processing for LLMs. Roughly 4 characters or 0.75 words in English.
Tool (MCP)
A capability exposed to agents through MCP. Examples: voice/transcribe, llm/chat, retrieval/search.
TPM (Tokens Per Minute)
A rate limit metric measuring allowed tokens per minute.
Transcription
Converting speech to text. See STT.
Transport
The communication protocol used between MCP clients and servers: HTTP/SSE, stdio, or WebSocket.
TTS (Text-to-Speech)
Converting written text to spoken audio. Also called speech synthesis.
TTFB (Time To First Byte)
Latency metric measuring time until the first byte of response is received. Critical for streaming.
U
Uptime
The percentage of time a service is operational. Calculated as (Total - Downtime) / Total × 100.
V
VAD (Voice Activity Detection)
Technology that detects when speech is present in audio, enabling natural turn-taking.
Vector Database
A database optimized for storing and querying embeddings. GateFlow uses pgvector.
Vector Store
See Vector Database.
Voice Mapping
GateFlow's system for mapping standard voice names to provider-specific voice IDs.
W
Warm-up
Pre-loading models or connections to reduce latency on first request.
WebSocket
A protocol for bidirectional real-time communication between client and server.
Webhook
An HTTP callback that delivers data to a URL when an event occurs.
Whisper
OpenAI's speech recognition model, available through GateFlow.
whoami
An MCP self-inspect tool that returns the current agent's identity and permissions.
Z
Zero-Downtime Migration
Switching models or configurations without service interruption, enabled by GateFlow's fallback system.
Acronym Quick Reference
| Acronym | Meaning |
|---|---|
| API | Application Programming Interface |
| ASR | Automatic Speech Recognition |
| BAA | Business Associate Agreement |
| FHIR | Fast Healthcare Interoperability Resources |
| GDPR | General Data Protection Regulation |
| GPAI | General Purpose AI |
| HIPAA | Health Insurance Portability and Accountability Act |
| HNSW | Hierarchical Navigable Small World |
| LLM | Large Language Model |
| MCP | Model Context Protocol |
| OCR | Optical Character Recognition |
| PHI | Protected Health Information |
| PII | Personally Identifiable Information |
| RAG | Retrieval-Augmented Generation |
| RLS | Row-Level Security |
| RPM | Requests Per Minute |
| SLA | Service Level Agreement |
| SSE | Server-Sent Events |
| STT | Speech-to-Text |
| TPM | Tokens Per Minute |
| TTS | Text-to-Speech |
| TTFB | Time To First Byte |
| VAD | Voice Activity Detection |
Next Steps
- FAQ - Common questions
- Getting Started - Start using GateFlow
- API Reference - Full API documentation